GAUGE THEORIES
*m
IN THE
TWENTIETH __CENTURY
GAUGE THEORIES
N p 4 IN THE
TWENTIETH CENTURY
EDITOR
J O H N C. TAYLOR
University of Cambridge
ICP
Imperial College Press
Published by Imperial College Press 57 Shelton Street Covent Garden London WC2H 9HE Distributed by World Scientific Publishing Co. Pte. Ltd. P O Box 128, Fairer Road, Singapore 912805 USA office: Suite IB, 1060 Main Street, River Edge, NJ 07661 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library CataloguinginPublication Data A catalogue record for this book is available from the British Library.
The editor and the publisher would like to thank the following for their assistance and their permission to reproduce the articles found in this volume: American Institute of Physics (JETP Lett.), American Physical Society (7. Math. Phys., Phys. Rev., Phys. Rev. Lett.), Elsevier Science Publishers B. V. (Nucl. Phys., Phys. Lett.), Kluwer Academic Publishers, Polish Academy of Sciences (Acta Phys. Pol.), Royal Society, SpringerVerlag (Z. Phys.)
GAUGE THEORIES IN THE TWENTIETH CENTURY Copyright © 2001 by Imperial College Press All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN ISBN
1860942814 1860942822 (pbk)
Printed in Singapore.
V
Contents
Preface Commentary 1 Gauge invariance in electromagnetism 1.1 J.E. Maxwell, A treatise on electricity and magnetism (1891), 3rd ed. (Dover, New York, 1954), article 616 1.2 H. Weyl, Spacetimematter, 4th ed. (1922), English translation by H.L. Brose (Dover, 1930), preface and para. 35 1.3 F. London, Quantenmechanische Deutung der Theorie von Weyl, Z. Phys. 42 (1927) 37589 1.3e English translation, Quantum mechanical interpretation of Weyl's theory . . . 1.4 O. Klein, Quantentheorie und fiinfdimensionale Relativitatstheorie, Z. Phys. 37 (1926) 895906 1.4e English translation, Quantum theory and fivedimensional theory of relativity 1.5 V. Fock, Uber die invariante Form der Wellen und der Bewegungsgleichungen fur einen geladenen Massenpunkt, Z. Phys. 39 (1927) 22632 1.5e English translation, On the invariant form of the wave equations and the equations of motion for a charged point mass 1.6 Y. Aharonov and D. Bohm, Significance of electromagnetic potentials in the quantum theory, Phys. Rev. 115 (1959) 48591 1.7 R.G. Chambers, Shift of an electron interference pattern by enclosed magnetic flux, Phys. Rev. Lett. 5 (1960) 35 Nonabelian gauge theories 2.1 C.N. Yang and R.L. Mills, Conservation of isotopic spin and isotopic gauge invariance, Phys. Rev. 96 (1954) 191201 2.2 R. Shaw, Invariance under general isotopic spin transformations, part II, chapter III of Cambridge PhD thesis (1955), pp. 3446 Gravity as a gauge theory 3.1 T.W.B. Kibble, Lorentz invariance and the gravitational field, J. Math. Phys. 2 (1961) 21221 3.2 C.N. Yang, Integral formalism for gauge fields, Phys. Rev. Lett. 33 (1974) 4457 Gauge invariance and superconductivity 4.1 Y. Nambu, Quasiparticles and gauge invariance in the theory of superconductivity, Phys. Rev. 117 (1960) 64863 4.2 B.D. Josephson, Possible new effects in superconductive tunnelling, Phys. Lett. 1 (1962) 2513 4.3 J. Schwinger, Gauge invariance and mass, Phys. Rev. 125 (1962) 3978 ix xi
1 3 38 53 64 76 84 91 95 102
2
105 110
3
119 129
4
132 148 151
. . .
vi 4.4 5 P.W. Anderson, Plasmons, gauge invariance, and mass, Phys. Rev. 130 (1963) 43942
153
Spontaneous symmetry breaking and particle physics 5.1 F. Englert and R. Brout, Broken symmetry and the mass of gauge vector mesons, Phys. Rev. Lett. 13 (1964) 3213 5.2 P.W. Higgs, Broken symmetries and the masses of gauge bosons, Phys. Rev. Lett. 13 (1964) 5089 5.3 G.S. Guralnik, C.R. Hagen and T.W.B. Kibble, Global conservation laws and massless particles, Phys. Rev. Lett. 13 (1964) 5857 5.4 T.W.B. Kibble, Symmetry breaking in nonAbelian gauge theories, Phys. Rev. 155 (1967) 155461 Gaugefixing in nonabelian gauge theories 6.1 R.P. Feynman, Quantum theory of gravitation, Acta Phys. Pol. 24 (1963) 697722 6.2 L.D. Faddeev and V.N. Popov, Feynman diagrams for the YangMills field, Phys. Lett. 25B (1967) 2930 6.3 G. 't Hooft, Renormalizable Lagrangians for massive YangMills fields, Nucl. Phys. B35 (1971) 16788 6.4 V.N. Gribov, Quantization of nonAbelian gauge theories, Nucl. Phys. B139 (1978) 119 Gauge identities and unitarity 7.1 C. Becchi, A. Rouet and R. Stora, Renormalizable models with broken symmetry, in "Renormalization Theory", ed. G. Velo and A.S. Wightman (D. Reidel, Dordrecht, 1976), pp. 26797 Asymptotic freedom 8.1 D.J. Gross and F. Wilczek, Ultraviolet behavior of nonAbelian gauge theories, Phys. Rev. Lett. 30 (1973) 13436 8.2 H.D. Politzer, Reliable perturbative results for strong interactions?, Phys. Rev. Lett. 30 (1973) 13469 Monopoles and vortex lines 9.1 P.A.M. Dirac, Quantized singularities in the electromagnetic field, Proc. Roy. Soc. A133 (1931) 6072 9.2 G. 't Hooft, Magnetic monopoles in unified gauge theories, Nucl. Phys. B79 (1974) 27684 9.3 A.M. Polyakov, Particle spectrum in quantum field theory, J E T P Lett. 20 (1974) 1945 9.4 H.B. Nielsen and P. Olesen, Vortexline models for dual strings, Nucl. Phys. B 6 1 (1973) 4561
157 160 162 165
6
173 199 201 223
7
242
8
271 274
9
278 291 300 302 319 334
10 Nonperturbative approaches 10.1 K.G. Wilson, Confinement of quarks, Phys. Rev. D10 (1974) 244559 . . . . 10.2 A.M. Polyakov, Compact gauge fields and the infrared catastrophe, Phys. Lett. 59B (1975) 824
Vll
10.3 G. 't Hooft, A planar diagram theory for strong interactions, Nucl. Phys. B72 (1974) 46173 11 Instantons and vacuum structure 11.1 A.A. Belavin, A.M. Polyakov, A.S. Schwartz and Yu.S. Tyupkin, Pseudoparticle solutions of the YangMills equations, Phys. Lett. 59B (1975) 857 11.2 B. Jackiw and C. Rebbi, Vacuum periodicity in a YangMills quantum theory, Phys. Rev. Lett. 37 (1976) 1725 11.3 C.G. Callan, Jr., R.F. Dashen and D.J. Gross, The structure of the gauge theory vacuum, Phys. Lett. 63B (1976) 33440 12 Threedimensional gauge fields and topological actions 12.1 A.N. Redlich, Gauge noninvariance and parity nonconservation of threedimensional fermions, Phys. Rev. Lett. 52 (1984) 1821 13 Gauge theories and mathematics 13.1 T.T. Wu and C.N. Yang, Concept of nonintegrable phase factors and global formulation of gauge fields, Phys. Rev. D12 (1975)384557
337
350 353 357
364
368
.
. I also thank Ron Shaw for information supplied to me. None of these people is to blame for deficiencies which remain in this volume. Sakura SchaferNameki and Fabian Wagner for translating the German papers (though they are not responsible for any errors introduced by my editing). Most of the ideas in the papers. so far as they have proved to be correct. such as [1]. The emphasis is on principles rather than applications. Tom Kibble.IX Preface My intention in this volume is to bring together some of the key papers in the development of gauge theories in physics. are to be found in modern textbooks. Inevitably. I am grateful to David Grellscheid. mainly from the 1920s to the 1980s. David Olive and Roger Phillips for their encouragement and advice. still less is it a textbook of physics. I am very grateful to Michael Atiyah. Some firsthand accounts of the history are to be found in [2]. The volume is not intended to be a serious work in the history of science. some equally important papers will have been omitted.
.
wrote the fivedimensional relativistic Schrodinger equation.3e). Schrodinger says. It is curious. a vector was assumed to undergo a change of length by a (pathdependent!) factor exp 7 / A • dx (1) where A^ is the 4vector of the electromagnetic potentials and 7 is some scalar coefficient." In 1927. that London is led to a sort of fivedimensional formalism. the word gauge (Eich in German) originates with Weyl's attempt [4] (an English translation appears in [3]) in 1918 to unify (classical) electromagnetism with Einstein's theory of gravity. Some of the papers reprinted here also appear in O'Raifeartaigh's book. In classical theory. with the metric components g5/J (/z ^ 5) 1 related to the electromagnetic potentials A^ (the interpretation of P55 is a moot point). The idea did not work. The other route to an appreciation of the role of gauge transformations in quantum theory began from another attempt to find a geometrical interpretation for electromagnetism. London transforms the electromagnetic potentials by an imaginary gauge transformation.4. on the basis of examples of periodic orbits. in view of Klein's work to be mentioned next. be identified with a length scale I.3. Maxwell's equations and the equations of motion of charged particles can be formulated in terms of the electric and magnetic fields. in modern terms. in 1926 (paper 1. As is well known. In 1922. in some sense. but this is confusing to modern readers.4e). . Fritz London published a "quantum mechanical interpretation of Weyl's theory" (paper 1. Paper 1. Schrodinger suggested that the constant 7 should be chosen to be imaginary.2 is a chapter from an English translation of Weyl's book SpaceTimeMatter (1922 edition). making a gauge transformation to get into the Coulomb gauge. Paper 1. under parallel transfer along a path C. to which I will make frequent reference. In his theory. "I do not dare to say whether this would make any sense in the context of Weyl geometry. Kaluza (English translation in [3]) proposed a fivedimensional spacetime. accompanied by a preface written in 1950 which explains how Weyl abandoned the theory. English translation 1. so that (1) becomes exp —le L A dx (2) An English translation of part of this paper is included in [3]. 2 and made the ansatz 0 is used instead of 5 by Kaluza. London argues that the wave function ip can.XI 1 Gauge invariance in electromagnetism The early history of gauge theories has been beautifully recounted by Lochlainn O'Raifeartaigh in The Dawning of Gauge Theory [3]. with proper time as the fifth coordinate. English translation 1. so gauge invariance plays no essential role (unless one wishes to write an action for the charged particle). transforming according to (2). In his very last equation. at least with the proviso that the path C in should be chosen to be the particle's path in a semiclassical approximation. But Maxwell in his Treatise did use the electromagnetic potentials.1 shows him. In 1921. What we would now call the KleinGordon equation — the nonrelativistic Schrodinger equation appeared only later in 1926. Oskar Klein.
The physical significance of the electromagnetic potentials in quantum theory was only appreciated in 1959.6) . 4 Shaw raises the question whether one of the gauge fields might be identified as the photon.2). Gauge transformations are just spacetimedependent transformations of the coordinate £5. Fock states very explicitly how the wave function behaves under gauge transformations. Something formally very similar was done independently by Fock and published in 1927 (paper 1.1). 2 Nonabelian gauge theories Nonabelian gauge theories were discovered by Yang and Mills. Both used modifications of the KaluzaKlein idea. and independently by Shaw (unpublished thesis. in Pauli's case. English translation 1. which in perturbative QED guarantees zero mass for the photon. first by Chambers in 1960 (paper 1. a sixdimensional spacetime. gauge invariance. But the authors had ambitious aims: the unification of electromagnetism with Einstein's theory of gravity. we are interested in these researches because they introduced gauge invariance into quantum theory.xii of periodic dependence on £5. Both Yang and Mills and Shaw were motivated by the idea of making isotopic spin invariance into a local.v{k) = 0. 4 Yang and Mills remark that no Ward identity analogous to k»Gy. At the end of Section 1. O'Raifeartaigh in [3] recounts the history in more detail. We now know 't Hooft's identity fcMfc"GMt. Both Yang and Mills and Shaw worried whether the spin 1 particles would have to have zero mass. who published their work in 1954 (paper 2.3). 3 . From a present day standpoint. both neutral and charged particles move classically on geodesies in fivedimensional spacetime. This amounts to compactifying the fifth dimension as a circle. by Ehrenberg and Siday [5]. In the KaluzaKlein approach.5. paper 2. but the geodesies followed by charged particles wrap around the £5 circle while those for neutral particles do not.3 An electron traversing paths outside a region of magnetic field can exhibit interference effects. = 0 would have the same effect (paper 6. Gauge transformations cannot in general make the phase factor (2) equal to 1 along all paths. and shows how both O. with. by Aharanov and Bohm (paper 1. In the nonabelian case.5e). had been proved. there is a generalization of the phase factor (2): P e x p _ig_ f A dx (3) h Jc A similar idea had appeared in a paper of 1949. He does this in the context of a now obsolete approximate strong interaction symmetry using the group 0{A). and the derivation of the wave equation of quantum theory. The electric charge is interpreted in terms of the angular momentum of the motion round the circle. The predicted effect was observed soon after. The Refractive Index in Electron Optics and the Principles of Dynamics.7). Klein (in 1938) and Pauli (in 1953) came close to discovering nonabelian gauge theories.
but uses general curvilinear coordinates uM. 4 Gauge invariance and superconductivity The role of gauge invariance in superconductivity was a matter of discussion for some years.1). but the quantities g^v ~ "itPkv may be interpreted as the metric of a Riemannian spacetime. He then goes on to regard the h^. he took the 10parameter group of inhomogeneous Lorentz transformations (Poincare group) as the group to gauge (that is. and an invariant action can be constructed therefrom (together with the determinant det/i^). the natural gauge theory is not Einstein's gravity. The first order. A paper of Yang in 1974 (paper 3.Xlll where P denotes the matrix ordering of the (Hermitian) matrix gauge potential A along the path C. Utiyama was working on a general theory of local gauge invariance [6] (reprinted in [3]). Yang does not introduce vierbeins. He defines *i& (3) where the xk are Cartesian coordinates. to make its parameters depend on position in spacetime). The similarity between gravity and YangMills theory was used by Feynman (paper 6. the (approximate) London equation VAj = . His approach was to gauge the sixparameter group of Lorentz transformation. Palatini form of gravity equations is obtained.1). In the phenomenological theory. Utiyama cast gravity as a gauge theory as follows. Prom his point of view. He began with a flat spacetime.2) also addresses this subject.1) looked at this question again. The gravitational field variables are the 16 vierbein components h£ and the 24 connection components AlJ[. as independent quantities. Kibble in 1960 (paper 3. The analogue of the field in YangMills theory is the curvature tensor.2 and 13. In these formulations. Unlike Utiyama. and shows that this requires the original space to be curved. Mills and Shaw. B the magnetic field and A the penetration depth) implies in general (although the original application was restricted to the transverse part of (5)) that j = ^(V^A) (5) . the analogy between YangMills theory and gravitation is certainly not complete: in YangMills there is no analogue of the vierbein. 3 Gravity as a gauge theory At about the same time as Yang.^ B (4) (with j the current. and he gauges the 16parameter group of general linear transformations. The quantity (3) is sometimes called the gauge phase factor (as in papers 3. Kibble works formally in a flat "background" spacetime.
Normally. A particularly bold modern exposition of this point of view can be found in Section 21. so as to include the longitudinal part.XIV for some scalar field <\>.2) is an example of the physical importance of the phase <p. the photon selfenergy must have the form (WPVMP2). Several authors (for example Nambu in paper 4. the phase difference depends linearly on time. (9) apparently vanishing6 at p2 = 0. II developed a pole at p2 = 0. but the Goldstone field is just the (j) in (5). In the GinzburgLandau [7] model for the superconducting phase transition. In the same year.1) and by Higgs (paper 5. in 1962 (paper 4. 6 In abelian theory. because of current conservation. Across a narrow insulating junction between two superconductors. 5 Spontaneous symmetry breaking and particle physics In 1964. by Englert and Brout (paper 5. and. or sometimes "vacuum symmetry breaking". In the relativistic case. and hence the current oscillates. The original BCS theory accounted for the transverse part of the current j in (6) only.pv term (9) makes no contribution. which is connected with longitudinal plasma oscillations (compression waves in the superconducting electrons). (6) (7) Much of the discussion hinged on the interpretation of the field <j).2). has no physical significance independently of the electromagnetic potential.6 of Weinberg's textbook [1]. (ft is the phase of the order parameter. for example. because of gauge invariance. by nonperturbative effects. gauge invariance implies that the photon must have zero mass. argued that the zero mass could be avoided if. a current flows which has a periodic dependence on the phase difference. cf> xf) + OJ . is now regarded as an example of spontaneous 5 breaking of gauge invariance. The fact that the condensate "chooses" a phase like this. The Josephson effect (paper 4. If a potential difference is maintained between the two superconductors. and 2ie<f> exp (8) is the (positiondependent) phase factor of this condensate. Anderson (paper 4. the p^.1) extended the theory in a gaugeinvariant way.3). under a gauge transformation where A > A + Vw. . 5 "Spontaneous symmetry breaking". Gauge invariance demands that. Schwinger. is said to occur when the ground state of a system does not have the invariance of the action which describes it. In the BCS microscopic theory.4) pointed out that just such an effect takes place in superconductors. The spontaneous symmetry breaking by the Cooper pair vacuum condensate might have been expected to entail a massless Goldstone boson. two short letters were published showing how gauge fields can get mass due to spontaneous symmetry breaking. there is a condensate of Cooper pairs.
1. Higgs draws attention to the partial multiplets of massive scalar particles. and no unique propagator. Englert and Brout refer to Nambu and JonaLasinio [8]. 7 . But this is not the subject of the present volume. which accompany the gauge particles. But this propagator has poles with four different polarization states: the two physical transverse ones. In the same year. although their field theories are essentially relativistic versions of this.XV They gave simple models in which some components of a multiplet of "elementary" scalar fields acquire a vacuum expectation value. even though they are scalars (or vectors in the case of gravity). A collection of original papers about dynamical gauge symmetry breaking appears in [11].3) discussed further the quantum theory of the spontaneous symmetry breaking. 7 This minus sign may be produced by making the ghost fields anticommuting quantities. The choice of component determines the symmetry breaking. Neither papers directly refer to the GinzburgLandau model of superconductivity [7]. he showed that the number of vector fields which acquire mass is equal to the number of symmetries which are broken. In QED. one needs to fix a gauge: otherwise there are no unique potentials. For example. now called Higgs particles. The other components of the scalar multiplet are swallowed by gauge fields. In particular they showed how Goldstone's theorem [9]. like B. Both sets of authors refer to previous work. In particular. one may add a "gaugefixing term" to the actions. Hagen and Kibble (paper 5. which shows that symmetry breaking requires massless quanta. In a paper of 1967. But in nonabelian theories this cancellation does not happen. Feynman analyzed perturbation theory diagrams. These are quanta of the fields which get vacuum expectation values. it is possible to show that the latter two unwanted poles give cancelling contributions (because the current is divergenceless). Electroweak theory [10] is based upon the breaking of of a gauged SU(2) x U(l) symmetry down to the electromagnetic C/(l) group. Kibble (paper 5. so that the photon propagator becomes r]^ being the Lorentz metric. undergo BoseEinstein condensation in the vacuum. 1964. 6 Gaugefixing in nonabelian gauge theories In order to calculate with quantum gauge field theories (at least by perturbation theory). giving three heavy vector particles and (at least) one Higgs scalar. and found he needed to include virtual "ghost" diagrams with an extra minus sign for each closed ghost loop. is evaded in the case of gauge theories. particularly related to superconductivity. de Witt). This was appreciated by Feynman in 1963 (reprinted in paper 6.4) gave a fuller account of symmetry breaking in nonabelian gauge theories. like fermion fields. providing the longitudinal polarization states necessary for spin 1 particles with mass. but also a longitudinal one (parallel to k) and a timelike one which has the wrong sign. together with a discussion involving other experts. that is. Guralnik. Higgs refers to Anderson.
Some wellknown textbooks begin their account of gauge theories by starting in such a gauge. and the Feynman integrals are not obviously welldefined. Another example of a physical gauge is to choose one of the Lorentz components of each gauge field to be zero. This introduces into perturbation theory denominators like 1/n • k (where n is a unit vector).4) observed that gaugefixing conditions. The renormalizability of field theories with massive vector mesons had been extensively investigated by M. and renormalizability is not obvious. Feynman was really concerned with quantum gravity. involving say d^A^ or diAi. Faddeev and Popov (paper 6. Veltman in the late 1960s. by manipulating the Feynman path integral. In QED. It is interesting to know whether there are choices of gaugefixing (necessarily not explicitly Lorentz invariant) in which the only dynamical fields are physical ones and in which a Hamiltonian exists. may not fix the gauge uniquely. Similar conclusions were reached by de Witt [12]. an example is the Coulomb gauge. but he used YangMills theory as a simpler model. The ghosts are necessary in order to cancel the Jacobian of the transformation from gauge variation to the gaugefixing function (which is trivial only in the abelian case). and from it derive covariant gauges (with ghosts).xvi These were necessary in order to complete the cancellation between longitudinal and timelike poles. so that unitarity is obvious. 't Hooft (paper 6. as discussed in the first place by Gribov in the paper reprinted here. where canonical quantization can be applied. In gauge theories with spontaneous symmetry breaking.2) gave a closedform derivation of the ghost contributions. Much of the importance of spontaneous symmetry breaking of nonabelian gauge theories is as a method to generate renormalizable theories of charged spin 1 particles (as required in electroweak physics). Perturbation theory is unaffected by this observation. In 1978. there is a question what types of gaugefixing are convenient to use. is an important question of principle. The methods then developed influenced the later work of 't Hooft (paper 6. One may use a physical (unitary) gauge in which the massive vector particles have propagators IJJXV MT k2 .8 In 1971.3) and 't Hooft and Veltman [16] on spontaneously broken gauge theories. see [13]. There may be "Gribov copies": different values of A related by gauge transformations and with the same value of the gaugefixing function. in which the longitudinal part of the potential is chosen to be zero and the timelike part is eliminated. 9 For more of 't Hooft's many important contributions to gauge theories. Gribov (paper 6.M2 + ie (11) But then the analysis of divergences by powercounting is complicated. and powercounting works just as in QED. But the generalization of the Coulomb gauge to nonabelian theories runs into complications and difficulties [14].3)9 introduced a class of gauges in which the 1/M 2 in (11) no longer appears. The existence of such physical gauges. 8 . but there are potentially very important nonperturbative effects. In 1967.
and. which is wellknown in QED. then Q2 = 0. BRST symmetry is nilpotent. independently in 1973 by Gross and Wilczek (paper 8. If the physical states 10 are assumed to be those annihilated by Q. The discovery of asymptotic freedom justified the use of perturbation theory in QCD. This symmetry is called BRST symmetry after the work of Becchi. 11 10 .1). It is essential that the shortdistance part should be insensitive to low energies.XV11 7 Gauge identities and unitarity In QED perturbation theory. The decisive discovery. The latter factorizes crosssections into a shortdistance part. the identities ensure that particles with equal bare charges also have equal renormalized charges. but indirect consequences of the original invariance remain. attempted to construct a gauge theory of strong interactions. because of isotopic spin invariance. but only up to terms suppressed by inverse powers of the energy [23]. gaugefixing destroys gauge invariance. and ghosts. of course. if Q is the operator which generates the symmetry. In QED.11 For unbroken nonabelian gauge theories (with not too many fermions or scalars) the opposite happens: the effective charge decreases at small distances (or at high energies). As an example.3. in analogy with condensed matter physics.2). This effect is sometimes explained intuitively as being due to "shielding" of the charge at larger distances by virtual charges in the vacuum. 17]. there is an effective charge. the corresponding identities are more complex. the experiments on highenergy electronnucleon scattering being done at SLAC. Renormalization must be done in a way consistent with them (although this must happen automatically if a gaugeinvariant regularization is used). and in most fourdimensional field theories. which explained all this. Sakurai. and are essential to the consistency. depending on distance. the idea of colour [20]. evidence from sum rules etc.12 As opposed to states containing arbitrary longitudinal and timelike polarizations. and by Politzer (paper 8. then unitarity is guaranteed. was that of asymptotic freedom. in the sense that. mix commuting quantities with anticommuting ones. this simple analogy is valid only for virtual particles which are bosons and are spinless [22]. in 1960 [18]. occurs also in QCD.1) and Tyutin. but more general gauge identities were soon derived [16. In fact. Rouet and Stora (see the article reprinted as paper 7. In nonabelian theories. Some of the key steps in the 1960s towards the theory of QCD were: the postulate of the existence (in some sense) of quarks [19]. given also a "factorization theorem". 12 The cancellation of infrared divergences. especially the unitarity of the calculation. The first such identities were derived by 't Hooft in paper 6. These identities express the invariance of the action (including the gaugefixing term) under certain nonlinear transformations which mix ordinary fields and ghost fields (that is. which increases at small distances (the bare charge being defined in the limit where the distance approaches zero). for the presence of free partons within hadrons [21]. to which perturbation theory is applicable. 8 Asymptotic freedom The inventors of nonabelian gauge theories had in mind the strong interactions. as in supersymmetry). and a longdistance part ("structure functions") which have to be taken from experiment. These consequences are the WardTakahashi identities [15].
compact and noncompact theories are quite different. the group is necessarily compact. 't HooftPolyakov monopoles are classical solutions (solitons) which must be stable for topological reasons. are magnetic monopoles (obeying Dirac's quantization condition (12).3) showed that. independently. These monopoles would be very heavy.2 is a discussion by Polyakov (1975) of the implications of compactness for confinement. or its covering group. or a numerical simulation. One of the most popular nonperturbative approaches has been to approximate the field theory by replacing spacetime (or possibly only space) by a discrete lattice. although no everywhere continuous magnetic vector potential A exists in the presence of a magnetic monopole. One of the seminal papers. Somewhat similar topological structures had been discovered in 1973 by Nielsen and Olesen (paper 9. the group may be taken to be U(l). Then an everywhere continuous wave function for the electrically charged particle does exist. where they overlap.1) proved that the existence of magnetic monopoles was consistent with quantum theory only if all pole strengths (m) and electric charges (e) satisfy the quantization condition em = . which is not. but they do exist in some broken grand unified theories (of strong and electroweak interactions). 't Hooft (paper 9. Wilson gives an example of a gaugeinvariant action on a lattice. They are field theory analogues of the quantized flux tubes in superconductors. is reprinted as paper 10. (12) The reason is that. by Wilson in 1974. Such monopole solutions do not exist in the minimum standard model. One may then use a strongcoupling approximation.2) and Polyakov (paper 9. In the case of QED. such as the explanation of confinement and the particle spectrum. For this to be a good approximation. 10 Nonperturbative approaches Asymptotic freedom justifies perturbative calculations of shortdistance effects in QCD. the lattice spacing must be sufficiently small. .XV111 9 Monopoles and vortex lines In 1931. Dirac (paper 9. which is compact.1. of course). however. In a nonabelian gauge theory based on a semisimple group. Among other things.4). are connected by gauge transformations w(x) satisfying exp {—) = 1. In 1974. but might have been produced during phase changes in the cooling early universe. These were stringlike. vector potentials do exist defined in different regions which. but longdistance effects. viewed from large distances. and proposes a mathematical criterion for confinement using the vacuumexpectation value of the Wilson loop operator (the operator in (3) with C a closed loop). indefinitely extended in one dimension. that is. In perturbation theory. in certain spontaneously broken nonabelian gauge theories.x integer . Paper 10. In lattice formulations. static classical solutions of the field equations exist which. are beyond its reach. this distinction has no force.
then.3) results from taking the gauge group to be SU(N) with N large (but Ng2 remaining finite). If. in fourdimensional Euclidean 13 field theory. They are topologically stable because the gauge function at infinity is nontrivially mapped onto the 3sphere at infinity. lepton. where 6 is a real number (called 9 by both groups of authors) which characterizes the theory (somewhat like another coupling strength).3). showing how an anomaly means that there need be no light isosinglet pseudoscalar particle (analogous to the pions) [24]. including. In QCD. In QCD. Such fields automatically obey the field equations. 11 Instantons and vacuum structure In 1975. the degree of the mapping being defined by an integer (where a is a group index). Belavin. exponentially suppressed. for a reason not fully understood. that is. instantons solve the UU(1) problem".XIX An alternative expansion to ordinary perturbation theory ('t Hooft. Instantons have a finite value of the (Euclidean) action. tunnelling probabilities. .1) discovered instantons. Dashen and Gross (paper 11. where F{2 = F34 etc. In electroweak theory. in particular.) YangMills fields whose potentials approach pure gauge values at infinity. and hence. selfdual or antiselfdual (meaning F* = ±F. 0 must be very small [25]. instantons predict tiny. field configurations with nonzero values of v are relevant to the Euclidean path integral. the effective value of g decreases with the relevant scale.2) and by Callan.and baryonnumber violating effects. how are they to be weighted? This question was answered in 1976 by Jackiw and Rebbi (paper 11. Polyakov. They must be weighted by a phase factor exp(iGv) (15) in the path integral. Schwartz and Tyupkin (paper 11. paper 10. The leading contribution comes from the sum of all planar graphs. But the phase (15) breaks P and CP invariance. The contribution to the path integral from field configurations near the instanton is proportional: 6XP \H*~) ' ( } Such contributions are beyond the reach of perturbation theory. "Euclidean field theory is relevant because properties of Lorentz field theory can be recovered from it. proportional to (13) and to 2 1/g (where g is the coupling strength). and the integral would vanish if A tended to zero at infinity faster than l/Vx*. Here the integrand is equal to a divergence. and so (14) is small for small instantons but not for large ones.
[4] H. [2] History of Original Ideas and Basic Discoveries in Particle Physics. Plenum Press. the base of the bundle is spacetime (or some part of it) and the gauge group acts on the fibre. but for gauge transformations which have a winding number w it changes by aw/g2. In applications to physics. Princeton University Press (1997). Weinberg. The Dawning of Gauge Theories. It is invariant under gauge transformations for which the gauge function maps trivially onto the sphere at infinity. and also to quantum field theory at very high temperatures. Weyl. they did not have the YangMills equations. and these constitute the holonomy group of the bundle. an anomaly introduces (Redlich.B. parity violating action (a is a constant) ^d3xeA^tr 0 ^X^txu 7J AxA^Av (16) This (ChernSimons) term [26] is topological.1) treats some physical situations using a mathematical point of view. Cambridge University Press (1996). .XX 12 Threedimensional gauge fields and topological actions In threedimensional 14 nonabelian gauge theories with an odd number of fermions. for instance in the case of a magnetic monopole. 1984. Princeton. Although mathematicians invented bundles. not long before the discovery by physicists of nonabelian gauge theories. edited by H. Sitzungber. The gauge potential is a connection. Preus. O'Raifeartaigh. Cambridge. A field is a section of a bundle. in the sense that no metric in the 3space is required in its definition.1) an effective. Newman and T. The condition that the exponential of i/h times (16) (occurring in the path integral) should change by a multiple of 2ir enforces a quantization condition on a. The gauge phase factor (3) is a parallel displacement. The Quantum Theory of Fields. [3] L. Some important mathematics about the classification of 4manifolds has been derived using these equations [27]. References [1] S. Akad. New York and London (1996). Berlin (1918) 465. Ypsilantis. "Threedimensional field theories are relevant to several situations in condensed matter physics. 13 Gauge theories and mathematics By a remarkable coincidence. Yang (paper 13. the relevant mathematical formalism — that of fibre bundles — had been invented by mathematicians. The mathematics becomes relevant when the bundle is not trivial. reprinted as paper 12.
B186 (1981) 376. Phys. Almqvist Forlag AB (1968) 367. [25] R. Phys.A.T. [15] J. Rev. Roy. Rev. [20] M. 40 (1978) 279. Jackiw. J.C. Lett. Ward. [24] G. Singapore (1994). Farhi and R. Rev. P. 6 (1957) 371. 10 (1972) 99. R. Lett.D.Y. GellMann. Phys. B44 (1972) 189. Svartholm. Phys. C.D. 122 (1961) 345. 10] S. Ginzburg and L. 197 (1987) 232. Weinberg. 101 (1956) 1597. Nucl. [27] S.L. Veltman.D. [6] R. Deser. 't Hooft. Goldstone. [22] R. Geom. [8] Y. Siday. 179 (1969) 1547. Quinn. Lett. Stockholm. Phys. 160 (1967) 1113. Bjorken. Rev. Theor. Under the Spell of the Gauge Principle. D22 (1980) 939. Nucl. S. 3155. Lett.J. [9] J. Nucl. Phys. 't Hooft. Phys. Rev. Phys. Phys.C. Peccei and H. Salam and S. Proc. 78 (1950) 182. de Witt. 40 (1978) 223. 139B (1965) 1006. (USSR) 20 (1950) 1064. 19 (1967) 1264. Nambu and G. Expt. Doria. Rev. Rev. B33 (1971) 436.P. Christ and T. [7] V.H. 22 (1968) 156. 23 (1969) 1415. Lett. in Elementary Particles Theory. Phys. World Scientific. Phys. CERN Report No. Vol. Ann. A. F. Fix. TH401 (1964). Feynman. D5 (1972) 3121. J. [23] R. 8 (1964) 214. Phys.E. Rev.C. Rev. [16] G.G.W. Weinberg. Rev. 12] B. Slavnov. [21] J. 13 (1964) 598. Rev. B. Phys. Han and Y. Lett. 3137. O. A. Phys. D16 (1977) 1791.E. 11] Dynamical Gauge Symmetry Breaking. Hughes. Lee and J. 162 (1967) 1195. Lett. Callan and D. World Scientific.J. [19] M. Templeton. 127 (1962) 965. edited by E. Phys. edited by N.W. Rev. Taylor. B50 (1972) 318. . Sakurai. Rev. Phys. [18] J. Utiyama.C. Nuovo Cim. Taylor. Jackiw and S. B168 (1980) 93. Diff. 19 of "Advanced series in mathematical physics". Taylor. Phys. Phys. 163 (1967) 1239. 18 (1983) 269. Phys. Rev. Phys. 't Hooft and M. i Mat. Nucl. Soc. 13] G. JonaLasinio. Zweig. ZinnJustin. Phys. Phys. Ann. Phys. Y. 11 (1960) 1. Lee. Greenberg. [17] J. Gross. B62 (1949) 8. D 1 4 (1976) 3432. Singapore (1982).XXI [5] W. and D7 (1972) 1049. G. 281 (2000) 409. A. Wilczek. [14] N. J. Ehrenberg and R. R. [26] S. Frenkel and J. Nambu. Phys. Rev. Phys. Takahashi. Rev. Rev. Doust and J. Salam. Weinberg. Lett. Donaldson. Phys. Landau. Teor.J.
.
the distinction which we have hitherto maintained between the magnetic force and the magnetic induction vanishes. 234. ^v=Tz4zw = /_• . I n order. however. they will discover nothing but electric circuits. 833. They may be combined so as to eliminate some of these quantities. (Dover. Art. and which is of very great importance.E.J 7fe' <1H / < .r d. and we shall find the magnetic force and the magnetic induction everywhere identical. According to Ampere's hypothesis. but our object at present is not to obtain compactness in the mathematical formulae.j J The components of the electric current Art. To eliminate a quantity which expresses a useful idea would be rather a loss than a gain in this stage of our enquiry. to be able to make use of the electrostatic or of the electromagnetic system of measurement at pleasure we shall retain the coefficient \x. the properties of what we call magnetized matter are due to molecular electric circuits. There is one result. [615. A Treatise on Electricity and Magnetism (1891). 3rd ed. however.F =Tz Tx' .] These may be regarded as the principal relations among the quantities we have been considering. dfi i _ j _ dy _ da d. GENERAL EQUATIONS.From: J. which we may obtain by combining equations (A) and (E). b u t to express every relation of which we have any knowledge. and if our mathematical methods are supposed capable of taking account of what goes on within the individual molecules. Article 616. 1954). because it is only in magnetized matter t h a t these quantities differ from each other. which will be explained in A r t . GO7. dG dF c = — — — d. dG  _dH "' ~ <l'/ . so t h a t it is only when we regard the substance in large masses t h a t our theory of magnetization is applicable.3 dy da J 1 re by equations (E). remembering t h a t its value is unity in the electromagnetic system.] The components of the magnetic induction are by equations (A). If we suppose that no magnets exist in the field except in the form of electric circuits. 501. 616. Maxwell. 615. New York.
+ =—h T . disappears from the equations (A). and it is not related to any physical phenomenon. omitting the accents. dJ n r.. 235 According to our hypothesis a. * The negative sign is employed here in order to make our expressions consistent with those in which Quaternions are employed. H=H' dy y (7) dz The quantity x. will give the true values of the components of 21. jj.W If we write G'= . c are identical with /xa. + dx .6i6. Similarly. If we suppose it to be zero everywhere.J VECTORPOTENTIAL OF CURRENTS.fi.(P ^dx2 d*_ dy2 d^s we may write equation (1).> dx dy dz .U = d2G _cPF dx dy dy2 d2F . (5) H '=Uff7d*d*d*> (6) where r is the distance of the given point from the element xy z. d2H dz2 dz dx (1) (2) (3) If we write and* 2 dF dG dH /T = =. 4 4 W ^ = •• _+v2e) dJ 2 dz + V # (4) IT\J. and equations (5). We therefore obtain 4 77 \S.fff^dxdydz. and the integrations are to be extended over all space. b. then j? = F' 4. dx dx G=G'+d^. / will also be zero everywhere. py respectively.
Relativity theory as expounded in this book deals with the spacetime aspect of classical physics. Of course. he would take into account certain events that have modified the situation in the intervening years. Thus. but ties them to the four v T . (1) The principle of general relativity had resulted above all in a new theory of the gravitational field. 35. a principle of gauge invariance in nature. with Einstein's gravitational potentials gik .. In the last two of its 36 sections. it proved insufficient to reach the goal at which classical field physics is aiming: a unified field theory deriving all forces of nature from one common structure of the world and one uniquely determined law of action. Brose (Dover. English transl. aside from the public's demand. as we know now. as I had assumed. my book describes an attempt to attain this goal by a new principle which I called gauge invariance (Eichinvarianz). 1930). may justify its reissue after so long a time. While it was not difficult to adapt also Maxwell's equations of the electromagnetic field to this principle. There holds. PREFACE TO T H E FIRST AMERICAN PRINTING HIS translation is made from the fourth edition of RAUM ZEIT MATERIE which was published in 1921.From: H. the book's contents are comparatively little affected by the stormy development of quantum physics during the last three decades. but it does not connect the electromagnetic potentials <pi . Weyl. This attempt has failed. This fact. H. SpaceTimeMatter (1922). had the author to rewrite the book today. I mention three such points. 4th ed. Preface and para.L.
1929. to be published soon). and my Rouse Ball lecture "Geometry and Physics" in Naturwissenschaften 19. They are all based on mathematical speculation and. pp. (2) Quite a number of unified field theories have sprung up in the meantime. Sitzungsber. Einstein. 1949. (3) A new development began for relativity theory after 1925 with its absorption into quantum physics. 2nd ed. GRUPPENTHEORIE UND QUANTENMECHANIK. SPACE^TIME STRUCTURE announced by Macmillan. Phys. gravitational and electronic. one could not have guessed this before the "electronic field" \p was discovered by quantum mechanics! Since then. 163171. Princeton. should encompass at least these three fields: electromagnetic. . the spinors. The most recent attempts by Schrodinger and by Einstein combine Eddington's idea of an affine field theory with that of dropping the requirement of symmetry for the metric tensor <7. 1950." in Proc. 1950. which introduced a new sort of quantities. H. 330. Weyl. Schrodinger. One nonspeculative development which deserves mention is Einstein's mixed metricafnne formulation in which both the gik and the TJ. N. Roy. 699701. Wissensch. 19 (1947/48).. Ultimately the wave fields of other elementary particles will have to be included too—unless quantum physics succeeds in interpreting them all as different quantum states of one particle. f. the article. p. as far as I can see. Kaluza's fivedimensional theory. pp. Leipzig 1928. Oct.vi components of the wave field \p by which Schrodinger and Dirac taught us to represent the electron. cf. 1925. compare my book. Preuss. a unitary field theory. Schrodinger's book. 52. For this and the following points. none has had a conspicuous success. "The Final Affine Field Laws. Einstein. 3rd ed. "Elektron und Gravitation" in Zeitschr. The first great success was scored by Dirac's quantum mechanical equations of the electron. (A) 51. Ak. J. 205216. THE MEANING OF RELATIVITY. Physik 56.t and the components r j ( of the affine connection. Of course. 1931. 1931. Review 77. 4958.. particularly in the garb of projective relativity. pp. and my lecture "50 Jahre Relativitatstheorie" at the first postwar meeting of the Gessellschaft deutscher Naturforscher und Aerzte (Munich. Irish Ac. A. Of the extensive literature I mention here only: E. so it seems to me. are taken as quantities capable of independent virtual variation. pp. however. p. 414. has been investigated and extended by several authors. Appendix I I .
vii besides the vectors and tensors into our physical theories. But difficulties of the gravest kind turned up when one passed from one electron or photon to the interaction among an indeterminate number of such particles. October 1950 HERMANN WEYL . Clarendon Press. Zurich.. 3rd ed. The generally relativistic formulation of these equations offered no serious difficulties. such as would account in the same basic manner for the elementary electric charge e as relativity theory and our present quantum mechanics account for c and h. 1947. THE PRINCIPLES OP QUANTUM MECHANICS. See Dirac's book. Oxford. In spite of several promising advances a final solution of this problem is not yet in sight and may well require a deep modification of the foundations of quantum mechanics.
This idea. is also dependent on a linear differential form <f>i dxi. In the latter case we have a static world that cannot exist without a masshorizon. namely. § 35. we then find that the metrical structure of the world. But this is incompatible with the intuitive principle of the relativity of motion. if facts are not to be violated drastically. only by maintaining the conception of infinitesimal parallel displacement of a vector set of axes ." let us assume that worldgeometry is of this kind. This principle could be satisfied. To be able to characterise the physical state of the world at a certain point of it by means of numbers we must not only refer the neighbourhood of this point to a coordinate system but we must also fix on certain units of measure. when applied to geometry and the conception of distance (in Chapter II) after the step from Euclidean to Riemann geometry had been taken. We wish to achieve just as fundamental a point of view with regard to this second circumstance as is secured for the first one. effected the final entrance into the realm of infinitesimal geometry. which we have treated more fully. In the former case the world would not be static as a whole. Eemoving every vestige of ideas of " action at a distance. which appears as a twodimensional configuration in the original coordinates is. therefore. it is the cylinder erected in the direction of the iaxis over the equator z — 0 of the sphere (65). the arbitrariness of the coordinate system. The question arises whether it is the first or the second coordinate system that serves to represent the whole world in a regular manner. assumed that uniform translation is a unique state of motion of a set of vector axes.282 THE GENERAL THEORY OF RELATIVITY indeterminate. Just as the step which led from the special to the general theory of relativity. so this extension affects immediately only the worldgeometrical foundation of physics. This edge. as also the special theory of relativity. but we found ourselves obliged to regard the * Vide note 33. and hence that the position of the axes at one moment determines their position in all other moments. is favoured by Einstein. by the Einstein Theory that was described in the preceding paragraph. de Sitter argues from this assumption (vide note 32). . besides being dependent on the quadratic form (1). threedimensional in the new coordinates. and the absence of matter in it would be in agreement with physical l a w s . Newtonian mechanics. this assumption. The Metrical Structure of the World as the Origin of Electromagnetic Phenomena * We now aim at a final synthesis.
M E T R I C A L S T R U C T U R E OF T H E W O R L D 283 affine relationship. The same certainty that characterises the relativity of motion accompanies the principle of the relatiYity of magnitude. it immediately suggests itself to us. for all the others statistical physics presents some reasonable argument which traces them back to the above two by the method of mean values. known to us besides those of gravitation and electromagnetic actions. . that gravitation is already contained in the guiding field besides inertia.* But. as something physically real that depends physically on the states of matter (" guiding field ")• The properties of gravitation known from experience. all physical fieldphenomena are expressions of the metrics of the world. Hence. we shall have to assign to the world besides its measuredetermination at every point also a metrical relationship. in spite of the existence of rigid bodies. but as being a phasefield having physical reality. inasmuch as this is true of the geodetic nulllines which are the determining factors in the propagation of light. this principle cannot be maintained without retaining the conception of infinitesimal congruent transformation . according to which the size of a body at one moment does not determine its size at another. unless we are to come into violent conflict with fundamental facts. which determines this displacement. finally. (Whereas the old view was that the fourdimensional metrical continuum is the scene of * It must be recalled in this connection that the spatial directionpicture which a pointeye with a given worldline receives at every moment from a given region of the world. not only to identify the coefficients of the quadratic groundform gucdxidx/c with the potentials of the gravitational field. We must not let our courage fail in maintaining this principle. We thus arrive at the inference : The world is a (3 + l)dimensional metrical manifold. however. that is. but also to identify the coefficients of the linear groundform <fodx» w i t h the electromagnetic potentials. And thus the general theory of relativity gained a significance which extended beyond its original important bearing on worldgeometry to a significance which is specifically physical. as we may call it. The electromagnetic field and tHe electromagnetic forces are then derived from the metrical structure of the world or the metrics. No other truly essential actions of forces are. teach us. depends only on the ratios of the gik'a. as the fact of the propagation of action and of the existence of rigid bodies leads us to found the affine relationship on the metrical character of the world which lies a grade lower. particularly the equality of inertial and gravitational mass. Now this is not to be regarded as revealing a " geometrical" property which belongs to the world as a form of phenomena.
is overcome in the new view. on which Maxwell's Theory is founded ! We have accordingly a good right to claim that the whole fund of experience which is crystallised in Maxwell's Theory weighs in favour of the worldmetrical nature of electricity.— — Hxk 7)Xi which is derived from it.if/af»da! . If our view of the nature of electricity is true. . things that exist " in " this world. The quantities of intensity are sharply distinguished from those of magnitude. but we must beware of letting this expression tempt us to form misleading pictures. is now to be identified with the distancecurvature file. however. the validity of which is wholly independent of whatever physical laws govern the series of values that the physical phasequantities actually run through. In this terminology the fundamental theorem of infinitesimal geometry states that the guiding field. is determined by the state of the sether.. . The electromagnetic fieldtensor. which is uniform and logical in itself. The linear groundform fcdxi is determined except for an additive total differential. According to Maxwell's Theory the same result obtains for the electromagnetic potential.) We shall use the phrase " s t a t e of the worldaether" as synonymous with the word "metrical structure. is free of arbitrariness. in the form of Action. but the tensor of distancecurvature /. (68) and it is just this one. _ Wi ^fa J ik — . and hence also gravitation. . then the first system of Maxwell's equations is an intrinsic law. and we must accept them in type and number in the form in which experience gives us cognition of t h e m : nothing further is to be " comprehended " of them.284 T H E G E N E R A L T H E O R Y OF R E L A T I V I T Y physical phenomena. Descartes' dream of a purely geometrical physics seems to be attaining fulfilment in a manner of which he could certainly have had no presentiment." in order to call attention to the character of reality appertaining to metrical structure . the physical essentialities themselves are. And since it is impossible to construct an integral invariant at all of such a simple structure in manifolds of more or less than four . which we denoted earlier by Fit. In a fourdimensional metrical manifold the simplest integral invariant that exists at all is fwx . The antithesis of " physical s t a t e " and " gravitation " which was enunciated in § 28 and was expressed in very clear terms by the division of Hamilton's Function into two parts.
sec . This particular form of the groundform (if we disregard quite particular cases) remains unaffected by a transformation of coordinates and a recalibration only if a. is equal to dx20 ~ (dxl + d*l + <&%)• . Hence. Thus." becomes intelligible through it. if the theory is to be in agreement with that atomistic structure of the world which. second) are to be chosen as arbitrary units . that is. The statical case occurs when the coordinate system and the calibration may be chosen so that the linear groundform becomes equal to <f>dx0 and the quadratic groundform becomes equal to fHx\ .1 . we have a threedimensional Eiemann space with the groundform do3 and two scalar fields in i t : the electrostatic potential <£. but there is not a factor of proportionality. and (f> has sec . in the statical ease. and not as one of the more general type. and the gravitational potential or the velocity of l i g h t / . the quantity Action is a pure number. In the linear groundform (f>idxi there is an arbitrary factor in the form of an additive total differential. and if the spacecoordinates are likewise transformed only among themselves.1 . The lengthunit and the timeunit (centimetre. as far as one may speak of a space at all in the general theory of relativity (namely.METRICAL STRUCTURE OF T H E W O R L D 285 dimensions the new point of view does not only lead to a deeper understanding of Maxwell's Theory but the fact that the world is fourdimensional. But this is only as it should be. We have the case of the special theory of relativity again.0 undergoes a linear transformation of its own. in the statical case). x3. we find dxl . it appears as a Riemann space. which has hitherto always been accepted as merely " accidental. but only on the spacecoordinates xv x2. If Xi. in which the transference of distances is found to be nonintegrable.{dx\ + dx\ + dxl) except for a factor of proportionality. whilst da2 is a definitely positive quadratic differential form in the three spacevariables. whilst the calibration ratio must be a constant. da2 has dimensions cm 2 .{dx\ + dx\ + dx2).rf<r2 whereby <f> a n d / a r e not dependent on the time x0. carries the greatest weight. then the transition from Xi to Xi is a conformal transformation. if the coordinates and the calibration may be chosen so that ds2 = dx% . / has dimensions cm . according to the most recent results (Quantum Theory). Xi denote two coordinate systems for which this normal form for ds2 may be obtained.
of the gaJs and their derivatives of the first and second order. on the one hand.286 T H E G E N E R A L T H E O R Y OF R E L A T I V I T Y The conformal transformations of the fourdimensional Minkowski world coincide with spherical transformations (vide note 34). in which <£.(x\ + x'i + x'!6).. respectively. W e choose as our initial physical law a Hamilton principle which states that the change in the Action I Yldx for every infinitely small variation of the metrical structure of the world that vanishes outside a finite region is zero. and Einstein assumed the Action to be an invariant with respect to transformations of the coordinates. in the form in which they hold in the special theory of relativity. We have here to add the further limitation that it must also be invariant with respect to the process of recalibration. used in § 28 (and which will only now be applied . with those transformations which convert every " sphere " of the world again into a sphere. According to Klein's method. g^ are replaced by <f>i . that is. that is. A OXi .— and Xgnc.u'x + u'i = 0. on the other hand.u\ . . Maxwell's equations of the aether. Hilbert. A sphere is represented by a linear homogeneous equation between the homogeneous " hexaspherical " coordinates _ (xx) + 1 (xx) ~ 1 where (xx) = x% . as expressed in the equation. and hence W is a scalardensity (in the true sense) which is derived from the metrical structure. The simplest example is given by Maxwell's density of action 1. built up. The spherical transformations therefore express themselves as those linear homogeneous transformations of the uja which leave this condition. They are bound by the condition u% . The Action is an invariant. . we must work out its implications. To test whether the new hypothesis about the nature of the electromagnetic field is able to account for phenomena. We assume that W is an expression of the second order..u'i «!. But we shall here carry out a general investigation without binding ourselves to any particular form of W at the beginning. invariant. (69) in which A is an arbitrary positive function of position. are therefore invariant not only with respect to the 10parameter group of the linear Lorentz transformations but also indeed with respect to the more comprehensive 15parameter group of spherical transformations (vide note 35). of the <£/s and their derivatives of the first order. Mie.
METRICAL STRUCTURE OF T H E WORLD 287 with full effect). In the normalisation the 8y*'s are components of a vectordensity (in the true sense).^ • • • (71) . according to (69). K is an infinitesimal scalarfield which characterises the event and which may be assigned arbitrarily. no objection to applying another normalisation in place of this one. if the Sfa's are regarded as the components of a covariant vector of weight zero and the Sgr^'s as the components of a tensor of weight unity. (There is. thus J s W d a . we express that IWda: is a calibration invariant. provided that it is invariant in the same sense.p)Zgafs + The 8Y*'S are defined uniquely by equation (70) only if the normalising condition that the coefficients (kia. which are valid for every scalardensity W which has its origin in the metrical structure. infinitely small increments 8^>i.) First of all. 8gik. {kia^Sg^i. * that is. we shall here deduce certain mathematical identities.i We indicate this by the formula JgoM = ^ ] . If we assign to the quantities <fu 9vc> which describe the metrical structure relative to a system of reference. As a result of this process. If the calibration ratio between the altered and the original calibration is A = 1 + ir. = p ^ W J ( w « f c + *TB^8^)<te . that it does not alter when the calibration of the world is altered infinitesimally. The 8Y*'S are linear combinations of ty„ S^andS^.. of course. Hi = . then the effect of partial integration is to separate the integral of the corresponding change SW in W over the region X into two parts: (a) a divergence integral and (b) an integral whose integrand is only a linear combination of 8<f>i and Sga. The w''s are components of a contravariant vectordensity. the fundamental quantities assume. but the W*'s are the components of a mixed tensordensity of the second order (in the true sense).0) be symmetrical in the indices k and i is added. and if X denote a finite region of the world. I. the following increments : % * = *gik. SY* = (ka)8<t>a + (ka. (70) whereby W*» = W1'*.
but choose for 7r a function which vanishes for points outside X but is > 0 throughout X. . The variation (70) of the integral of Action must vanish on account of its calibration invariance. X For a given scalarfield IT it holds for every finite region X. say positive. for a particular .rW*) • • • • { > If we substitute (72) in this. If we choose this region for X in (73). let the following expressions result: B*(«r)=.288 T H E G E N E R A L T H E O R Y OF RELATIVITY If we substitute these values in Sv*.s* +  ^ . and observe that. h * « . that. Now that this has been ascertained. since the ~—'s are the components of a covariant vectorfield which is derived from the scalarfield. then the first integral vanishes. and consequently we must have 3(B*(7T) cW* . s* is a vectordensity. we see that (73) gives J p)(B*'& ^ U ~dXk =0 . It further follows from this. in the manner familiar in the calculus of variations. we may write. (72) They are the components of a vectordensity which depends on the scalarfield T in a lineardifferential manner. but the second is found to be positive—which contradicts equation (73). . J— L X ^H^****)**0 X • (73) . . (74) This immediately gives the identity ~f +W =0 . that is.a. If the function of position on the left were different from 0 at a point Xi. instead of the preceding equation. then it would be possible to mark off a neighbourhood X of this point so small that this function would be positive at every point within X. . we have X X If we transform the first term of the second integral by means of partial integration. and h*" is a contravariant tensordensity of the second order.
that is. Accordingly. express the process of congruent transference of a distance. I n view of the skewsymmetry of h the first is a result of the second. h** is a linear tensordensity of the second order.*gik l)Xk 7)Xi 7)Xr ) in which tv denotes an infinitesimal scalarfield that has still been left arbitrary by our conventions. if we remain at the same spacetime point. then this single formula resolves into the identities: According to the third identity. (21'). (77) If we wish to express the invariance with respect to the coordinates alone we must make IT = 0 .) = .(<£. (71) (76) ~ t>(Jik = \ 0i>— + Qkr — + ^— f' I . in which each point undergoes a displacement whose components are £}. S'Z = 0. let the metrical structure accompany the deformation without being changed. in (76) we must choose w not equal to zero but equal to . We subject the worldcontinuum to an infinitesimal deformation. x . namely. in fact.l(4>£). arbitrary values may be assigned to 77.0 . Then. by (20). but indicates that 87 = . ^—^— . 8' the change in the same quantity if we share in the displacement of the spacetime point. signifies that the deformation is to make the two groundforms vary in such a way that the measure I of a lineelement remains unchanged.£') if we are to arrive at invariant formulae. . — .METRICAL STRUCTURE OF T H E W O R L D 289 point. however. This convention. This equation does not.l^ih'x. since I I . but the resulting formulae of variation (76) have not then an invariant character. Let 8 signify the change occasioned by the deformation in a quantity. The invariance of the Action with respect to transformation of coordinates and change of calibration is expressed in the formula which relates to this variation : 8' * Wrfx = rpmi) + s w j dx .
•k The equation then becomes . we get an integral. pi + f p . The fact that S*(£) is a vectordensity dependent on the vectorfield $( expresses most simply and most fully the character of invariance possessed by the coefficients which occur in the expression for S*(f). /?). . which was discussed just above. it follows from this that the Si's are not components of a mixed tensordensity of the second order: we call them the components of a " pseudotensordensity ". particularly in the case of the second equation (78). If we insert in (77) the expressions (70) and (78). S*(£) is a vectordensity which depends in a linear differential manner on the arbitrary vectorfield £'.S#it = $ik + faNow that the calibration invariance has been applied in I. For the variation (78) let W£* + SY* = S*(fl. whose integrand is On account of — + g*P<f>i — !"„. we may in the case of (76) restrict ourselves to the choice of w. and which we found to be alone possible from the point of view of invariance. We write in an explicit form S*(*)~S** + Bft£+*H* 7>X l a TjXabXfi (the last coefficient is. if we introduce the mixed tensor <*£*' i i 1)x. in particular. ai 7>xi and of the symmetry of W " ' we find <£ + ^ ) w " =r^ W* = riw. too.290 T H E G E N E R A L T H E O R Y OF RELATIVITY The change in the two groundforms which it represents is one that makes the metrical structure appear carried along unchanged by the deformation and every lineelement to be transferred congruently. The invariant character is easily recognised analytically. symmetrical in the indices a. of course.
W f o = Dxk 0 . . S* = IS* . and the quantities H = 0. on account of the conditions of symmetry W = 0.= 0. . iXy Hf + H?*° + H f = 0 If from (4) we replace in (3) Hf" by . (79) 3(B*(fl . Consequently g' = 0. Wdx J = 0. . . we get ft>(B*(fl .Hf"» we get that Hf .^ i ! l . . ].METRICAL STRUCTURE OF T H E WORLD 291 If we apply partial integration to the last member of the integrand. . ft.—In the case of Maxwell's Actiondensity we have.//»f*«..Hf i)xr is skewsymmetrical in the indices a.K?y . we get ^Hf? = Q Example. If we introduce H f in place of H" we see that (3) and (4) are merely statements regarding symmetry. but (2) becomes Sj + ^ _ + J H ' = Wf • • • (81) d) follows from this because. that is.WfeQ J 2>xic dx + fr . (*™J _ riWt) and + faW = 0 . According to the method of inference used above we get from this the identities : [ . . (80) The latter resolves into the following four identities : dx* a ~i)Xk ^Xa (Hf + H? ) + £ ^ i . hik = f*. as is immediately obvious 3Y* = P*8(fc.
According to (752) the electromagnetic laws have the following form : ^ = s« [and (67)] .292 T H E G E N E R A L T H E O R Y OF R E L A T I V I T Y Hence our identities lead to oxa oxi M l We arrived at the last two formulas by calculation earlier. we get h l i = fik. (82) in full agreement with Maxwell's Theory. 8 = 0. Thus there are among the fieldequations five superfluous ones corresponding to the transition (dependent on five arbitrary functions) from one system of reference to another. S* = 0. the former on page 230. Field L a w s and Theorems of Conservation. is the Action. in (70). outside it. The particular form of Hamilton's function W affects only the formulas which state that current and fielddensity are determined by the phasequantities <f>i. and for X we take the whole world or a region such that. . . Without specialising the Action at all we can read off the whole structure of Maxwell's Theory from the calibration invariance alone. Of these. we see from this that the following invariant laws are contained in Hamilton's Principle : YP = 0 W* = 0. which have been stated in (74) and (79). If iWda. Just as the s*'s constitute the density of the 4current. which is as it should be. s* is the density of the 4current. and the linear tensordensity of the second order h'* is the electromagnetic density of field. which is valid only in empty space. the latter was found to express the desired connection between Maxwell's tensordensity Sk of the fieldenergy and the ponderomotive force. Between the lefthand sides of these equations there are five identities. g^ of the aether. we take for 8 an arbitrary variation which vanishes outside a finite region. we have to call the former the electromagnetic laws. so the scheme of S f s is to be interpreted as the pseudotensordensity of . In the case of Maxwell's Theory in the restricted sense ( W = 1).—If. the latter on page 167. we get [sWdx = f(w*'8<fc + Wk8gik)dx. the latter the gravitational laws.
they follow in two ways from Ss* Dw the field laws. not their derivatives . logically. The field laws and their accompanying laws of conservation may. For . This second component L contains only the (/it's themselves. in consequence of the . the calibrationinvariance here appears as such. . each part had to be treated differently. To these four laws there is to be added the law of conservation of electricity. Earlier we derived the law of conservation of energymomentum from the coordinateinvariance only owing to the fact that Hamilton's function consists of two parts. § 28) we introduced W* as the energy components. For the gravitational component (W = G) we defined the energy by means of *Sj (§ 33). . and. and.£WJ. but for the electromagnetic component (W = L. but also to . According to (75^ and (80^ t h e theorems of conservation are generally v a l i d .METRICAL STRUCTURE OF THE WORLD 293 the energy. are distinguished by a prefixed asterisk. W = 1. the actionfunction of the gravitational field and that of the " physical phase " . The form of the gravitational equations is given by (81). and — is not only identically equal to T — ! . ^*Scoordinateinvariance.r — = 0 lXk are generally valid. If those quantities. be summarised conveniently in the two equations »tfw = 0 *m _ =0 Attention has already been directed above to the intimate connection between the laws of conservation of the energymomentum and the coordinateinvariance. But the *Sf s are not the energymomentum components of the twofold actionfunction which have been used as a basis since § 28. indeed.— is not only identically equal to ^—. In the simplest case. for a quantity of this kind we have. by (75) and (80). be a property of invariance which will introduce a fifth arbitrary function ./itW*. this explanation becomes identical with that of Maxwell. then. instead of from (78). . and the component results had to be combined appropriately (S 33). but also — to r^oW^ . which are derived from W^" + Sv* by taking the variation of the fundamental quantities from (76) for the case ir — 0. by (802).. Wj = S* Hence (if we use the transformations . the "theorems of conservation" . corresponding to it. there must.
Just as is the case in § 33 in which we integrate over the crosssection of a canal of the system. they coincide with the fluxes . namely. of course. if the si?s and S f s vanish outside the canal. and the other does not contain the derivatives of the ga's. These discrepancies are removed only here since it is the new theory which first furnishes us with an explanation of the current s*. as the flux of a certain spatial field through a surface O that encloses the system. as a result of the laws of conservation. we can adapt the two different definitions of energy to one another although we cannot reconcile them entirely. The laws of conservation of the s*'s and the S / s are then likewise not bound by an assumption concerning the composition of the Action. Thus. accordingly. by Maxwell's equations (82) and the gravitational equations (81). even if the field has a real singularity within the canal of the system. (81) (in which the righthand sides are to be replaced by zero) in terms of the quantities h and H belonging to the altered field. let us replace this field within the canal in any arbitrary way (preserving. and let us define the s''s and the S*'s by the equations (82). If we regard this representation as a definition. carry along the metrical structure and the lineelements " unchanged" in our sense and not in that of Einstein. of the electromagnetic density of field h*. the integral theorems of conservation hold. a continuous connection with the region outside it) by a regular field. The integrals of these fictitious quantities s° and S°. What is done by Einstein's theory of gravitation with respect to the equality of inertial and gravitational matter. that it recognises their identity as necessary but not as a consequence of an undiscovered law of physical nature. on the other hand. and of the energy S*. of which the one does not contain the fa's and their derivatives. are constant. the system has a constant charge e and a constant energymomentum / . after the total energy had been introduced in § 33. we have once again passed beyond the stand taken in § 28 to a point of view which gives a more compact survey of the whole. is accomplished by the present theory with respect to the facts that find expression in the structure of Maxwell's equations and the laws of conservation. Both may be represented. so we find here that. which is no longer bound by the assumption that the Action is composed of two parts.294 THE GENERAL THEORY OF RELATIVITY which the fundamental quantities undergo during an infinitesimal alteration of the calibration). which are to be taken over the crosssection of the canal (the interior of O). The virtual deformation of the worldcontinuum which leads to the definition of S* must. To prove this.
and omit the divergence which vanishes when we integrate over the world. then we may assume that the grit's and the <£. If we use the formula (61). * Vide note 3G. The simplest assumption * for purposes of calculation (I do not insist that it is realised in nature) i s : W . We assume that .) By means of the substitution Xi = ix'i we introduce coordinates of the order of magnitude in general use (that is having dimensions comparable with those of the human body). £ is a very small constant. The g^'s do not change during this transformation. § 36. and we get Y = G + al + \ Jg{l .. of course.\F> Jg + al . It follows that 8W = . agrees with Einstein's Theory. we choose the coordinates Xi so that points of the world whose coordinates differ by amounts of the order of magnitude 1. in addition. The Fundamental Equations of Mechanics We have now to show that if we uphold our new theory it is possible to make an assumption about W which. . (62). then our principle of action takes the form B\Ydx = 0.Fis positive. (83) The quantity Action is thus to be composed of the volume. If. are separated by cosmic distances. a fact that the potentials vary perceptibly by amounts that aie extraordinarily small in comparison with cosmic distances. as far as the results that have been confirmed in experience are concerned. § 17 for F. by means of partial integration. the calibration may then be uniquely determined by the postulate F = .1 . . we convert the worldintegral of 8($R Jg) into the integral of SG (§ 28). and if. since on O the imagined field coincides with the real one. (It is.'s are of the order of magnitude 1. the positive constant a is a pure number. § 17) and of Maxwell's action of the electromagnetic field.iFS(Fjg) + ±F*S Jg + a81.3 (<^)} (84) This normalisation denotes that we are measuring with cosmic measuring rods. measured in terms of the radius of curvature of the world as unit of length (cf.S I M P L E S T PRINCIPLE OF A C T I O N 295 mentioned above over the surface fi. thus SW = the variation of iF Jg + \ Jg + al. . Application of the Simplest Principle of Action.
a cosmological term that is dependent on the electromagnetic potentials occurs. whereas in Einstein's Theory (cf. Since. But. y occurs in place of a. let it be e in ordinary electrostatic units.«2 .gap). Einstein's cosmological term must be 3 _ supplemented by the further term . but if we take Xi to represent the coordinates previously denoted by x'i. if we use these units. Hence our theory necessarily gives us Einstein's cosmological term <. we have that the mass present in the world determines the curvature. too. then ¥ = ( G + al) + ^g{l~ 3(4.X \fg.296 THE GENERAL THEORY OF RELATIVITY if we simultaneously perform the recalibration which multiplies ds'2 by 5. in (84).nX Jg(<j>i<l>1)'.{f} + h2m* + %fr . It seems to the author that just this is what makes Einstein's cosmology physically possible. § 34) there must be a preestablished harmony between the universal physical constant X that occurs in it. we have . in our ordinary measures. and the total mass of the earth (because each of these quantities in themselves already determine the curvature of the world). and if Vrik are the components of the affine relationship corresponding to these coordinates.4% rik . the radius of curvature of the world. here (where X denotes merely the curvature). we arrive exactly at the classical MaxwellEinstein theory of electricity and gravitation. In the case in which a physical field is present. In the new system of reference we then have g'ih = gtk. Our theory is founded on a definite unit of electricity. F' = .is accordingly. If g^. by neglecting the exceedingly small cosmological terms. To make the expression correspond exactly with that of § 34 we must set jr = X. The uniform distribution of electrically neutral matter at rest over the whole of (spherical) space is thus a state of equilibrium which is compatible with our law. <f>i retain their old significance. $i = «^i. Thus. and in the components r£t of the gravitational field.
on the one hand.21 SIMPLEST PRINCIPLE OF ACTION 297 our unit is that quantity of electricity whose gravitational radius is 4 J. The cosmological factor which Einstein added to his theory later is part of ours from the very beginning. Variation of the <£/s gives us Maxwell's equations. in this case. • (86) This follows.<M*. times the radius of curvature of the world. as is to be expected S? = at + {G + ikJgtf ..R + 2Wite — a!. and. The conservation of electricity is expressed iri the divergence equation 0. From the equation 8' I Ydx = 0 for a variation 8' which is produced by the displacement in the true sense [from formula (76) with £*' = const. from Maxwell's equations. of cosmic dimensions.— 4>i V0.F = 2X again gives (86). 7r = 0]. therefore. on the other hand. We actually find. that B + 2 \ = f (<fr<^) and this in conjunction with . we get . but must. we have simply 3X . Variatio of the git's gives the gravitational equations BH where * . Just as according to Maxwell the aether is the seat of energy and mass so we obtain here an electric charge (plus current) diffused thinly throughout the world.Sl = .* ^ f GM}. be derivable from the gravitational equations according to our general results. like the quantum of action 1.Oj • (85) T? = {l + MrS')}8?•Mkr . I t is. We get for the pseudotensordensity of the energymomentum. by contracting the latter equations with respect to ik.
The answer is in the affirmative. according to our earlier remarks. that is. For if we recalibrate the statical world in accordance with the postulate F — .(&£'). write Maxwell's equations in the form ~dXi then set IT = .Fga. add it to (87). g* = .it. We then get. for this calibration. after multiplying the resulting equation by a. MSfrO _ Q ~bxh The following terms occur in S*: the Maxwell energydensity of the electromagnetic field the gravitational energy GSf . = — (i = 1. * where we set F. If. and. in order to deter . we must. 3) oxi > T? y* = .^ G QXi «M and the supplementary cosmological terms The statical world is by its own nature calibrated. however. it follows that F = const.1 and distinguish the resulting quantities by a horizontal bar. we get <j>i = B • Jp. The particle is necessarily charged electrically.2. in fact. From the fact that a further electrical term becomes added to Einstein's cosmological term.298 where T H E G E N E R A L T H E O R Y OF RELATIVITY To obtain the conservation theorems. The question arises whether F = const. the existence of a material particle becomes possible without a mass horizon becoming necessary. J~g = F1 Jg and equation (86) gives 3 aFi /bXi From this.
and hence h% is uniquely determined. and <$>. and / 4= 0 . . ^h*$hr 4 . by multiplying the differential equation of <p by <f>.. w.is infinite to the second order. as a result of the free choice of the unit of time. k of / and 4> we have / o = j^S</>5If diametral points are to be identified. I n / and <f>.2 —this proves. then the quantities that occur as functions of z = V r j . is (the accent denotes differentiation with respect to r). that 0 .S I M P L E S T PRINCIPLE OF A C T I O N 299 mine the radially symmetrical solutions for the statical case. The differential equations themselves show that the development of hz. and integrating from 0 to r 0 . a common constant factor remains arbitrary (a circumstance that may be used to reduce the order of the problem by 1). h*di*\ . A. have a singularity at least ultimately when r = 0. cj> must be an even function of z. A to the first order. <> f. If the equator of the space is reached when r = r 0 . Por otherwise it would follow.•—whereas for the initial values of / 0 . then the integral whose variation must vanish. if we let r decrease from r 0 . respectively.according to powers of z begins with the term hi. kr" / . we again use the old terms of § 31. incidentally. a r'W* Variation of VA'") ~~ 2 V " A " ' As a result of the normalisations that have been performed. which satisfy the given conditions {vide note 37). but must. and take $ to mean the electrostatic potential.rmust exhibit the following behaviour for z = 0 : / and <p are regular. leads to the equations AA' = . the spatial coordinate system is fixed except for a Euclidean rotation. and the solution is uniquely determined by the initial values for « = 0. where ° Xrjl . It cannot remain regular in the whole region 0 < J r < ^ r 0 . /i. that X must be positive (the curvature F 2 negative) and that r% > .
T* = /*«. to show that in general the inertial mass is the flux of the gravitational field through a surface which encloses the particle. The fact that the phasequantities vary appreciably in regions whose linear dimensions are very small in comparison with —^ may be explained. namely (§ 2 7 ) : dmds — fxdx. according to the former equation. In doing this we are. one of weight 0. and no external forces act on it. as they contradict the postulated properties of invariance. This is obviously intimately connected with the circumstance that the integral Irfmrfs has now no meaning at all. Prom this we see at once that we must start from the laws of conservation . And we see that these initial conditions are impossible in the new theory for the same reason as in Einstein's Theory. according to the latter. In conclusion. We took the first step towards giving a real proof of the mechanical equations in § 33. as was mentioned at the end of § 33. perhaps. is a tensordensity in the true sense. p is a scalardensity of weight £. We shall also take this opportunity of carrying out the intention stated in § 32. according to which the charge and the mass are determined from infinity.«* are quite impossible here. and hence cannot be introduced as " substanceaction of gravitation ". because they lead to a false value for the mass. There we considered the special case in which the body is completely isolated. For.300 T H E G E N E R A L T H E O R Y OF R E L A T I V I T Y Matter is accordingly a true singularity of the field. we shall set up the mechanical equations that govern the motion of a material particle. we shall now endeavour to make good this omission. of course. so to speak. this agrees with the idea developed in § 32. A The fact that all elementary particles of matter have the same charge and the same mass seems to be due to the circumstance that they are all embedded in the same world (of the same radius r 0 ) . debarred from using a substance which is in motion. the hypotheses corresponding to the latter idea. since T. namely. and. In actual fact we have not yet derived these equations in a form which is admissible from the point of view of the general theory of relativity. even when the matter has to be regarded as a singularity which limits the field and lies. by the circumstance that a value must be taken for r% which is enormously great in comparison with . outside it.. that is.
in accordance with a remark at the end of § 35. be defined in the above manner. the integral sum at over the three last terms. The antithesis of kinetic and potential which receives expression in the fundamental law of mechanics does not.^ . whose dimensions are great compared with the actual essential nucleus of the particle. and which are functions of the time alone. as a flux through the surface of Cl. represent the energy (i = 0) and the momentum (i = 1. indeed. In this way we arrive at the mechanical equations rl T wKi < 90 > On the left side we have the components of the " inertial force. indeed. however. be marked off around the material particle.SIMPLEST PRINCIPLE OF ACTION 301 which hold for the total energy. over CI. be such that the spaces x0 = const. over Cl. the first member (k — 0) gives the timederivative . xs." and on the right the components of the external "fieldforce". Let a volume Cl. Let the coordinate system consisting of the "timecoordinate" x0 = t and the "spacecoordinates" xv x." used at the end of § 35. If we integrate the equation (89) in the space x0 = const. 2.2. 3) of the material particle. Not only the fieldforce but also the fourdimensional momentum Ji may be represented. In the course of the motion Cl describes a canal in the world. and have nothing to do with the singular states or phases in its interior. It is of fundamental importance to notice that in them only such quantities are brought into relationship with one another as are determined by the course of the field outside the particle (on the surface of Cl). intersect the canal (the crosssection is the volume CI mentioned above). but small compared with those dimensions of the external field which alter appreciably. The integrals J S °idx1dxidx3 = Ji which are to be taken in a space x0 = const. becomes transformed by Gauss' Theorem into an integral E{ which is to be taken over the surface of Cl. conditioned by the resolution into space . leads to the mechanical equations proved above. in the interior of which the current filament of the material particle flows along. If the interior of the canal encloses a real singularity of the field the momentum must. depend actually on the separation of energymomentum into one part belong'ng to the external field and another belonging to the particle (as we pictured it in § 25). and then the device of the " fictitious field. but rather on this juxtaposition.
To do this. it gives the flux e (and not 0.£ of (91) is that derived from the potential . of course. We should be reverting to the notion of substance if we were to interpret the MaxwellLorentz equation °J— = Pu l so literally as to apply it to the volumeelements of an electron. as it would be in the case of a solution of (91) which is free from singularities) of the electric field through an envelope O enclosing the particle. these properties are not lost when an arbitrary solution /. If we discuss Maxwell's equations first. such a one is given by /& = const. if we introduce at the moment under consideration a coordinate system in which the electron is at rest. But it is clear that this assumption comes into question only for quasistationary motion.* of equations (91). of the first and the three last members of the divergence equations which make up the laws of conservation. which deals with " Force and I n e r t i a " (vide note 38). but are very limited in three other dimensions. This stand was taken most definitely by Mie in the third part of his epochmaking Foundations of a Theory of Matter. justified only when we are dealing with quasistationary motion.'602 THE GENERAL THEORY OF RELATIVITY and time.. that is. The field w h i c h surrounds the m o v i n g electron m u s t be of the t y p e : /. . Its true meaning is rather this : Outside the flcanal. The term pu1 in Lorentz's equation is to express the general effect of the chargesingularities for a region that contains many electrons. This assumption concerning the constitution of the field outside 12 is. Our next object is to work out the full consequences of this view for the principle of action adopted in this chapter. free from singularities. Nothing at all can be asserted about what happens during rapid acceleration. that is. it is necessary to ascertain exactly the meaning of the electromagnetic and the gravitational equations. we may disregard gravitation entirely and take the point of view presented by the special theory of relativity. is added to fa.* + fa. when the worldline of the particle deviates by a sufficiently small amount from a straight line. (91) The only statical radially symmetrical solution /. On account of the linearity of equations (91). the homogeneous equations ^=0hold . . . The opinion which is so . on the circumstance that the singularity canals of the material particles have an infinite extension in only one dimension.
our fieldequations make assertions only about the possible states of the field. The above assumption about the singular component / of the field surrounding the particle is. to be smoothed out. of course. It is justified only if Lorentz's equations are interpreted in the too literal fashion repudiated above. also. We assume that this solution is characteristic of the moving particle in the following sense: We consider the values traversed by the gik's outside the canal to be extended over the canal. which the path of the material particle cuts out in the metrical picture of the world. it is assumed that the constitution of the electron is not modified by the acceleration. as it should be if the solution were free from singularities. a greatly accelerated particle emits radiation. which depends on a single constant m. the mass. according to Bohr. at t h a t point. work out other assumptions.SIMPLEST PRINCIPLE OF ACTION 303 generally current among physicists nowadays. is quite different from that imagined by Hertz. Let ds be the corresponding propertime differential. The flux of a gravitational field through a sufficiently great sphere described about the centre is not equal to 0. the particle is a radiating atom.= dx% (dx\ + dx% + dx%) . If. seems to the author quite unfounded. We may. we shall for the present adopt the point of view of the original Einstein Theory. (This is a possible state of the field which is caused by matter in a manner which. This gap is filled by the Quantum Theory in a manner of which the underlying principle is not yet fully grasped. ds. and that they may move permanently in these orbits without emitting radiations. but equal to m. In it the (homogeneous) gravitational equations have (according to § 31) a statical radially symmetrical solution. true for a quasistationary electron.) As far as gravitation is concerned. Bohr's Theory of the Atom has led to the idea that there are individual stationary orbits for the electrons circulating in the atom. by supposing the narrow deep furrow. For a point of the streamfilament we may introduce a ( " n o r m a l " ) coordinate system such that. that. and if. according to classical electrodynamics. only when an electron jumps from one stationary orbit to another is the energy that is lost by the atom emitted as electromagnetic energy of vibration (vide note 39). and by treating the streamfilament of the particle as a line in this smoothedout metrical field. in our opinion. If matter is to be regarded as a boundarysingularity of the field. for example. the fn/s will have to be represented as the field of an oscillating Hertzian dipole. and not about the conditioning of the states of the field by the matter.
rr = 0 . provided the distance from the particle is sufficiently great. as we know from § 32. I n terms of these coordinates the field is to be expressed by the abovementioned statical solution (only.dx„ = e* l 2 3 *xk J in any arbitrary coordinate system . If we regard the normal coordinates xt as Cartesian coordinates in a fourdimensional Euclidean space. if . that ?_. then the picture of the worldline of the particle becomes a definite curve in the Euclidean space. that is. once and for all. the calibration normalised by F = const.) To return to the derivation of the mechanical equations! We shall use. Our assumption is. Consequently. on the surface of O). the equation of definition gives ^\oi ds^ us. admissible again only if the motion is quasistationary.304 T H E G E N E R A L T H E O R Y OF RELATIVITY the derivatives . of course. the arguments set dxl at . this assumption is legitimate only in the quasistationary case. by fusing them into a continuum . To prove this.. a linear problem .dx<. Since (if we neglect the cosmologioal terms) the s''s on this surface vanish.— = 0 is integrated. moreover. and the direction of the streamfilament is given by dx0: dxl: dx2: dx3 = 1 : 0 : 0 : 0 . But this assumption is compatible with the fieldlaws only if e = const. The determination of the electromagnetic field is then. from which the canal of the particle is to be cut out). \B°dx. (The transformation of the homogeneous gravitational equations into nonhomogeneous ones. e* is independent of the choice of the fictitious field. we shall deduce from a fictitious field that fills the canal regularly and that links up with the really existing field outside. due to the presence of masses. of course. = s f . if we base our calculations on the normal coordinate system. it is to have the form fa + fa mentioned above (with fy = const. to be neglected in comparison with the influence of the mass.. we may assume the gravitational field to be that mentioned above.^ i vanish. The influence of the charge of the electron on the gravitational field is. inasmuch as it may be represented as a fieldflux through the surface of O. in a certain neighbourhood of the worldpoint under consideration. on the right side of which the tensor /xUiUk appears. if this picturecurve is only slightly curved at the point under consideration. as in the gravitational case. and we shall neglect the cosmologioal terms outside the canal. takes account of the singularities.
The last term is to be neglected since it contains the weak field /"as a square. the first term contributes nothing. we sacrificed the coordinate invariance of our Action. moreover. because. we define aS* by means of (Rf . By integrating (92) we get (90). with regard to the representation of the energymomentum components by means of fieldfluxes. since the flux of a constant vector through a closed surface is 0. by applying the process of partial integration to arrive at (84). left with only the electromagnetic part. which is to be calculated along the lines of Maxwell. But this gives us £»• = «/of . each of them is composed of three terms in accordance with the formula (/ + /) 2 =/ 2 + 2yy + 72. whereby n Ki expresses itself as the fieldflux through the surface Q. In these expressions the fictitious field may be replaced by the real one./Wb. we must notice at once that. Since the components of Maxwell's energydensity depend quadratically on the field / + /. and. Hence we must proceed as follows. With the help of the fictitious field which bridges the canal regularly.i$R) + (G8f The equation i^<***} PA* = 0 (92) da* is an identity for it. the representation of e* as a fieldflux shows that e* = e.(Bji8*iR)byl8j. we may replace . If we use the normal coordinate system the part due to the gravitational energy drops out. Passing on from the charge to the momentum. we may not refer to the general theory of § 35. therefore. thel middle term alone remains. in accordance with the gravitational equations. for its components depend not only linearly but also quadratically on the (vanishing) derivatives •—. If we use the normal coordinate system at one point. In the case of each.SIMPLEST PRINCIPLE OF ACTION 805 out in § 33 show that e* is independent of the coordinate system chosen.• uXi We are.
by using identities (92) and treating the crosssection of the streamfilament as infinitely small in comparison with the external field) (1) that. we find that not only at the point of the canal under consideration.* . according to the fieldlaws. (93. which is to fit into the remaining part of the field. for which the gravitational field that surrounds the particle has the form calculated in § 31. but also just before it and just after it Jt = mui («> = ^ J . . and in the immediate neighbourhood of which the field has the required structure. .• • / " • * " • (94) For the relations (94) are invariant with respect to coordinate transformations. In the normal coordinate system.306 T H E G E N E R A L T H E O R Y OF RELATIVITY Concerning the momentumquantities we see (in the same way as in § 33. is that the quantities e and'm that characterise the singularity at each point of the canal remain constant along the canal. . a necessary condition for a singularity canal.„* ds ox1 m In the light of these considerations. according to page 272 : Jx = J3 = J 3 = 0. Hence.* ^ « ^ . The 0th of these equations gives us : sr = 0. since the fictitious field may be chosen as a statical one. On account of the property of covariance possessed by J. however. we find that. the <7". But in any arbitrary coordinate system we h a v e : ^ . but with a charge of the coordinate system in the canal) the quantities J* retain their values. for coordinate transformations that are to be regarded as linear in the crosssection of the canal. and hence = TO. not with this. and (2) that if we alter the fictitious field occupying the canal (in § 33 we were concerned.. but that the worlddirection of the canal satisfy the equations ^ _ i * & W = !. Hence the equations of motion of our particle expressed in the normal coordinate system are * £ > . thus the field equations require that the mass be constant. and agree with (93) in the case of the normal coordinate system./„. it seems to the author that the opinion expressed in § 25 stating that mass and fieldenergy are .'s are the covariant components of a vector which is independent of the coordinate system. and J0 = the flux of a spatial vectordensity through the surface of CI.
and to ascribe to the world relationships such as obtain in Einstein's Cylindrical World (jj 34). that similar atoms radiate the same frequency measured in the proper time ds corresponding to the normalisation F — const. the new standpoint in no wise signifies a relapse to the old idea of substance. a radiating particle loses inertial mass of exactly the same amount as the electromagnetic energy that it emits. when there are cut out of it canals of circular crosssection which stretch to infinity in both directions. the worlddistance traversed by it during a period were to be transferred congruently from period to period in the sense of our worldgeometry. certainly not take place in the way described. but also that mass as the point of attack of the metrical field is identical in nature with mass as the generator of the metrical field. and the whole of Mie's view of matter assumes a fantastic. a natural result of the special theory of relativity that we should come to this conclusion.. an atom) with infinitely small period. but it deprives of meaning the problem of the cohesive pressure that holds the charge of the electron together. different periods when they meet at a later worldpoint B. It is only when we arrive at the general theory that we find it possible to represent the mass as a fieldflux. is by no means as convincing as in the theory of Einstein : it loses its validity altogether if a principle of action other than that here discussed holds. independently of their previous * The invariant quadratic form F. This view of m states not only that inertial and gravitational masses are identical in nature. in general. With about the same reasonableness as is possessed by Einstein's Theory we may conclude from our results that a clock in quasistationary motion indicates the proper time Iris which corresponds to the normalisation F = const.) This may be proved simply and rigorously from our present point of view. unreal complexion. .SIMPLEST PRINCIPLE OF ACTION 307 identical is a premature inference. then two clocks which set out from the same worldpoint A with the same period.g. which traverse congruent worlddistances in A during their first period will have. of course. (In this example Einstein first recognised the intimate relationship between energy and inertia.* If during the motion of a clock (e. It was. ds* is very far from being distinguished from all other forms of the type E. That which is physically important in the statement that energy has inertia still persists in spite of this. that is. The orbital motion of the electrons in the atom can. For example.1) as is the ds 3 of Einstein's Theory. Moreover. therefore.246). ds 2 (E being a scalar of weight . which does not contain the derivatives of the potentials at all. For this reason the inference made in our calculation of the displacement towards t h e infrared (P.
will manifest any arbitrary deviations in the positions of the axe's. Its direction is determined at every moment. this does not explain why an electron itself after an arbitrarily long time still has the same charge. But even if this be the case. independently of the state of the system at other moments. and why this charge is the same for all electrons. This circumstance shows that the charge is determined not by persistence but by adjustment: there can be only one state of equilibrium of negative electricity. Diametrically opposed to this is the case of a magnet needle in the magnetic field. for example. for it . for example. in virtue of its constitution. The same reason enables us to draw the same conclusion for the spectral lines of the atoms. for what is common to atoms emitting equal frequencies is their constitution and not the equality of their frequencies at some moment when they were together far back in time. but once this arbitrary initial direction has been fixed the direction of the axis of the top when left to itself is determined from it for all time by a tendency of persistence which is active from one moment to another. What is the source of this discrepancy between the conception of congruent transference and the behaviour of measuring rods. adjusts itself to the field in which it is embedded. and which meet after the lapse of a great length of time. following the tendency of persistence. Maxwell's equations for the charge e of an electron make necessary de the equation of conservation 57 = 0. whereas for a congruent transference it dl would have to satisfy the equation r. nevertheless two tops which set out from the same point with axes in the same position.I.308 T H E G E N E R A L T H E O R Y OF R E L A T I V I T Y histories. In the same way. as. since they can never be fully removed from all influences. There is no a priori ground for supposing a pure transference. Thus although. and atoms ? We may distinguish two modes of determining a quantity in nature. at each instant the axis experiences an infinitesimal parallel displacement. $. clocks. that of persistence and that of adjustment. to which the corpuscle adjusts itself afresh at every moment. We may prescribe to the axis of a rotating top any arbitrary direction in space. Neither does a measuring rod at rest in a statical field undergo a congruent transference. namely. for the measure I = do2 of a measuring rod at rest does not alter. This difference is illustrated in the following example. since the atoms emit spectral lines of definite frequencies. by the fact that the system. for rotations of the top in Euclidean space. the length of a measuring rod is determined by adjustment. to be integrable. obviously. = .
that it has decided advantages so far as the deeper problems. One of these is Maxwell's: * = */*/* (95) another is the F~ used just above.2 formed in a perfectly rational manner from the components of curvature. if what we assumed above is true for direction then at each moment of the motion of the top the rotation vector would experience a parallel displacement. But its trace £ is a scalar—of weight .) We may briefly summarise as follows: The affine and metrical relationship is an a priori datum telling us how vectors and lengths alter. According to the same law by which (95). But curvature is by its nature a linear matrixtensor of the second order: FitdiTiSa. The two quantities L and I seem to be invariant and of the kind sought.*. (Perhaps the time of rotation of a top gives us an example of a timelength that is determined by persistence. We have seen that it accounts equally well for all the phenomena which are explained by the latter theory and. The worldcurvature makes it theoretically possible to determine a length by adjustment. invariants . and in what proportion persistence and adjustment modify one another. Nevertheless.S I M P L E S T PRINCIPLE O F A C T I O N 309 would be impossible to give to this rod at t h i s point of the field any length. Only four of these invariants may be set up. in the way that I can prescribe its direction arbitrarily.e. compatible with the new axiom of calibration invariance. from which every other may be built up linearly by means of numerical coefficients (vide note 40). which most nearly approaches the MaxwellEinstein theory. from the principle of action. can be found only by starting from the physical laws that hold. say two or three times as great as the one that it now has. But to what extent this is the case in nature. is produced from the distancecurvature/a. We may certainly assume that W has the form Wjg. and they can be formed most naturally from the curvature. if t h e y happen to follow the tendency of persistence. in which W is an invariant of weight . i. The subject of the above discussion is the principle of action. I doubt whether the Hamiltonian function (83) corresponds to reality. The multiplication is in this case to be interpreted as a composition of matrices . the square of the numerical value. such as the cosmological problems and that of matter are concerned. indeed.we may form iFttF* (96) from the total curvature. (96) is therefore itself again a matrix. In consequence of its constitution the rod assumes a length which has such and such a value in relation to the radius of curvature of the world.2.
310 THE GENERAL THEORY OF RELATIVITY of this natural and simple type. here we may expect to meet with new results. but it was also pointed out what obstacles lie in the way of such an irrational Action. There seem to be two ways out of this difficulty. but also of those favoured here. and hik = f**. It seems more probable that W is a linear combination of L and I. But in the question of the mechanical equations and of the relationship holding between the results obtained by measuringrods and clocks on the one hand and the quadratic form on the other. It is quite a different matter to inquire into the excitation or cause of the fieldphases that appear to be possible according to these laws. One serious objection may be raised against the theory in its present state: it does not account for the inequality of positive and negative electricity (vide note 42). in the discussion on Mie's theory. there is the following view which seems to the author to give a truer statement of reality. Or. it directs our attention to the reality lying beyond the field. Either we must introduce into the law of action a square root or some other irrationality. We have here occupied ourselves only with the field which satisfies certain generally invariant functional laws. Calculations by Pauli (vide note 41) have indeed disclosed that the field determined in § 31 is not only a rigorous solution of Einstein's equations. But we saw above (§ 33) that the sign. which emits energy owing to the jump of an electron from one orbit to another in accordance with Bohr's hypothesis. Maxwell's equations become then as above : (when the calibration has been normalised by F = const. The gravitational laws in the statical case here. This oneness of sense in Time exists beyond doubt—it is. Thus in the aether there may exist convergent as well as divergent electromagnetic waves.) 8' = a constant multiple of Jgft. it was mentioned how the desired inequality could be caused in this way. too. indeed. too. secondly. exist only in a fourdimensional world at all. but only the latter event can be brought about by an atom. . agree to a first approximation with Newton's laws. This example shows (what is immediately obvious from other considerations) that the idea of causation (in contradistinction to functional relation) is intimately connected with the unique direction of progress characteristic of Time. namely P a s t » Future. so that the amount by which the perihelion of Mercury's orbit advances and the amount of the deflection of light rays owing to the proximity of the sun at least do not conflict with these equations. situated at the centre. indeed. the connecting link with the old theory seems to be lost. the most fundamental fact of our perception of Time—but a priori reasons exclude it from playing a part in physics of the field.
but now the rigid motionless character has become transformed into one which gently yields and adapts itself. of course.S I M P L E S T PRINCIPLE OF A C T I O N 311 of an isolated system is fully determined. The laws of the metrical field deal less with reality itself than with the shadowlike extended medium that serves as a link between material things. it plays a part that is in no wise different from that which space with its rigid Euclidean metrical structure plays. Examples of such regularities of structure that concern. through the quantum theory. This connects the inequality of positive and negative electricity with the inequality of Past and F u t u r e . and of natural forces. but the roots of this problem are not in the field. and physics would no longer be far from the goal of giving so complete a grasp of the nature of the physical world. leading from the Euclidean metrical structure to the mobile metrical field which . by the facts of the quantum theory. For the present. of itself. that logical necessity would extract from this insight the unique laws that underlie the occurrence of physical events. of matter. not the field. but the problem of matter is still wrapt in deepest gloom. If Mie's view were correct. We must here state in unmistakable language that physics at its present stage can in no wise be regarded as lending support to the belief that there is a causality of physical nature which is founded on rigorously exact laws. powerless.Future. Whoever looks back over the ground that has been traversed. however. But even if we recognise the limited range of field physics. but lie outside it. it seems that the t h e o r y of statistics plays a part in it which is fundamentally necessary. merely provisional in character. But freedom of action in the world is no more restricted by the rigorous laws of field physics than it is by the validity of the laws of Euclidean geometry according to the usual view. we must gratefully acknowledge the insight to which it has helped us. we must reject these bold hopes. according to the old view. has already reached a deeper stratum of reality than is accessible to field physics. but the causes of the fieldphases are instanced : by the existence of cylindrically shaped boundaries of the field : by our assumptions above concerning the constitution of the field in their immediate neighbourhood : lastly. we could recognise the field as objective reality. Statistical physics. The extended field. Past s." is merely the transmitter of effects and is. Nevertheless. "eether. and above all. But the way in which these regularities have hitherto been formulated are. and with the formal constitution of this medium that gives it the power of transmitting effects. as soon as a definite sense of flow. has been prescribed to the worldcanal swept out by the system.
in spite of all disappointments and errors. must be overwhelmed by a feeling of freedom won—the mind has cast off the fetters which have held it captive. it is yet able to follow the intelligence which has planned the world. and that the consciousness of each one of us is the centre at which the One Light and Life of Truth comprehends itself in Phenomena. and which includes the field phenomena of gravitation and electromagnetism.312 T H E G E N E R A L T H E O R Y OF R E L A T I V I T Y depends on matter. but that. . makeshift in the struggle for existence. Our ears have caught a few of the fundamental chords from that harmony of the spheres of which Pythagoras and Kepler once dreamed. a too human. H e must feel transfused with the conviction that reason is not only a human. whoever endeavours to get a complete survey of what could be represented only successively and fitted into an articulate manifold.
v. Note 38. 93144). d. p. p. the articles of Abraham quoted in Note 4 . also G. G. d." Sitzungsber. (310). 3. Physik. Bd. 31. Pauli. Bd. pp. This theorem was proved by Liouville: Note IV in the appendix to G.. Zeitsohr. p. also Weyl. Wissensch. Society). Note 42. 504. vol. Preuss. A similar tendency is displayed (although obscure to the present author in essential points) in E . 21 (1919). Bd. Note 40. pp.. Zeitsohr. Akad. d. Note 41. (275). This fact. Cf. vide Picard. 21. (273).. pp. d. Traite d'Analyse. t. 22364. Bd. Amsterdam. 29th. 135. Weitzenbock in a letter to the present author. MerkurPerihelbewegung und Strahlenablenkung in Weyl'sGravitationstheorie. No. Note 36. in Wien. p. Monge. 142. Hilbert (I. Physik. Concerning other attempts to derive Electricity and Gravitation from a common root cf. d. p. Physik. 609. Note 34.36). 1917. Ges. Zeitsohr. Note 37. The theory contained in the two following articles were developed by Weyl in the Note "Gravitation und Elektrizitat.. Atombau and Spektrallmien. d. Einstein. 2 Mitt. Nordstrom. also W. 45767.e. Bateman. 1917 6. Sitzungsber. Ann.. (286). d. 34956. Pauli (I. Astronom. Proc. p. 20 (1919). v. (299). This was proved by R.e. Cunningham. his investigation will appear soon in the Sitzungsber. Concerning such existence theorems at a point of singularity. Weyl. Ann. d. Klein (I. 15 (1914).301. Physik. Akad. 7 (Dec. Preuss. As described in the book by Sommerfeld. (295).. Akad. 1919.e. Note 30. Eine neue Erweiterung der Relativitatstheorie. Ann. Die Gravitation als elektrodynamische Erscheinung. (280).. Society (2). Physik. Bd. which here appears as a selfevident result. Bd. idem. (282). (276). 1918. Note 33. also Ann. (286). 20 (1919). Note 32. 39 (1913). Verhandl. Zur Theorie der Gravitation und der Elektrizitat von H.. Reichenbacher (Grundziige zu einer Theorie der Elektrizitat und Gravitation. Cf. vol xx. Application de l'analyse a la geom^trie (1850). 63 (1920). pp. p. d. d. E. 7799. Cf. d. d. Akad."). (309). Preuss. Weknsch. On the mass of a material system according to the Theory of Einstein. (282). (310). d. also F . had been previously noted : E.. Wieehert. 1917). (1910). 52 [1917]. Weyl. 63 [1920]. H. Bd. Ann.. 1919 and 1921. de Sitter's Mitteilungen im Versl. Wissensch. d. Akad. Physik. 59 (1919). Deutschen physik.'). Akad. Cf. Wissensch. viii. 465. p. of the London Mathem. pp. Einstein arrived at partly similar results by means of a further modification of his gravitational equations in his essay : Spielen Gravitationsfelder im Aufbau der materiellen Elementarteilchen eine weaentliche Rolle ? Sitzungsber. Pauli. as also his series of concise articles : On Einstein's theory of gravitation and its astronomical consequences (Monthly Notices of the E. (303). Vieweg. Note 35. te Amsterdam. (302). Bd. Note 31. 742. Physik. d. Note 89. W. Note 29. Physik. p. Nordstrom. Wetensch. Wissensch. .324 BIBLIOGRAPHY Note 28.
dafi nur die V e r h a l t n i s s e der gih an e i n e r Stelle. Von P. sinngemaB festgelegt werden konnen. hat bekanntlich kiirzlich durch W e y l eine auflerordentlich schone und einfache Vervollstandigung erfahren. § 1. 15. Oder. II. 1927. am 18. auch einen vorliiufigen zusammenfassenden Bericht in Naturwiss. dafi der Mafistab. Kap. Um dieser Aussage R i e m a n n s einen Sinn zu geben.375 Quantenmechanische Deutung der Theorie von Weyl 1 ). Demgegeniiber macht W e y l mit Recht geltend. dafi die Annahme eines solchen starren Mafistabes einer radikalen Nahgeometrie zuwider sei. und dementsprechend setzt er fiir die Anderung dl eines EichmaBstabes von der Lange I bei einer infinitesimalen Verschiebung d x{ an: dl = Itpidx'. Ges. III. (Eingegangen am 25. Man kann den R i e m a n n s c h e n Raumbegriff betrachten als die Aufhebung des Vorurteils.) Kap. Quantenmechanische iTmdeutung der Theorie von Weyl. Nichtintegrabilitat schlie'Bt Eindeutigkeit nicht aus. Stuttgart. dafi die K r i i m m u n g s v e r h a l t n i s s e an e i n e r Stelle des Raumes verbindlich sein miifiten fiir die Kriimmung an a l i e n anderen. ein „starrer" Mafistab sei. wenn man (1) integriert: l = lgeln^ (2 ) 1 ) Vorgetragen teilweise auf der Tagung des Gauvereins Wurttemberg der D. K a p i t e l I. 187. Dezember 1926. Die Undulationsmechanik von de Broglie und die Theorie von Weyl. Die Idee einer „reinen Nahgeometrie". Charakteristika der Mafiverhiiltnisse des Raumes — ahnlich den gik. welcher an jeder Stelle zur Bestimmung der Koeffizienten gik der metrischen Fundamentalform ds2 = gikdxidxk zur Anwendung gelangt. Februar 1927. (1) wobei die Proportionalitatsfaktoren cpt Funktionen des Ortes sind. London in Stuttgart. § 2. Die Theorie von Weyl. war zunachst die A n n a h m e notwendig. Die Identitat von tp und Weyls Eichstrecke. nicht ihre Absolutbetrage. zuerst von R i e m a n n konzipiert. . D i e T h e o r i e von W e y l . vgl. Phys. Kap. I.
(?0 = I am Anfang der Verschiebung). es sei denn. kann man laut ihrer Definition (3) die Identitat aussprechen (die Dimensionenzahl der Mannigfaltigkeit sei 4): Die formale Ubereinstimmung dieser vier 'Gleichungen mit dem einen System der M a x w e l l s c h e n Gleichungen rot © f — £ = 0.des elektromagnetischen Viererpotentials. c div £> = 0. Uhren) zu denken. !Q. charakterisiert durch die V a r i a b i l i t a t des E i c h m a B es. reprasentieren Mafistabe. die (pi seien bis auf einen konstanten Proportionalitatsfaktor zu identifizieren mit den Komponenten <Pt. die fik entsprechend mit den elektromagnetischen Feldstarken (£. ebenfalls als eine Eigenschaft der Maflverhaltnisse des Raumes. London. K) J^~~drt versehwinden. mit welcher W e y l allein auf Grund dieser ganz formalen Zuordnung seine Lehre von der eichgeometrischen Deutung des Elektromagnetismus aufgespiirt h a t : In der Gravitationstheorie war es eine p h y s i k a l i s c h e T a t s a c h e .376 F. das Prinzip der Aquivalenz zwischen trager und schwerer Masse. deren U n a b h a n g i g k e i t von der V o r g e s c h i c h t e durch die Scharfe der Spektrallinien belegt ist. (2 a) Man wird die ungeheure Kuhnheit bewundern. im Widerspruch zu dem nicht integrablen . Es ist also zu schreiben: Z= Z0 ea J ** dXi (« = Proportionalitatsfaktor). In der Theorie der Elektrizitat dagegen war eine solche Tatsache nicht bekannt: Es bestand keine Veranlassung. an einen universellen Einflufi des elektromagnetischen Feldes auf die sogenannten starren Mafistabe (bzw. das e l e k t r o m a g n e t i s c h e F e l d . tjber diese GroBen fik. sowie einige weitere formale Analogienhaben W e y l zu dem Schlufi gefiihrt. B. Ganz im Gegenteil. die Atomuhren z. welche E i n s t e i n zu seiner geometrischen Deutung anregte. Das EichmaB ist im allgemeinen vom Wege abhiingig (nicht integrabel). In folgerichtiger Erganzung der geometrischen Deutung der G r a v i t a t i o n durch die v a r i a b l e n K r i i m m u n g e n des R i e m a n n s c h e n Raumes dachte sich W e y l den noch ubrigbleibenden Teil physikalischer Wirkungen. dafl die GroBen tik _dg>t dcpk „.
mit einem ganzzahligen Multiplum der P l a n c k schen Konstanten als Periode. E r hielt an seiner Auffassung test und entzog die eben geschilderten Widerspriiche der Diskussion durch eine etwas dunkle Umdeutung des Begrifi's „realer MaBangabe". Ich werde vielmehr zeigen. daB die Natur von diesen schonen ihr gebotenen geometrischen Moglichkeiten Gebrauch machen miisse. dafi ip eine eindeutige Funktion des Raumes. wie es Zeitsohrift fUr Physik. Auf diese abstrakte Ausgestaltung der Theorie brauche ich niclit einzugehen. welches W e y l im magnetischen Felde annimmt. die W e y l solchen elementarsten Erfahrungen zum Trotz niclit von dem Gedanken abgehen liefi. dafi gerade der pragnanten urspriinglichen Fassung der Weylschen Theorie eine noch viel groCere Spannkraft innewohnt.Quantenmechanische Deutung der Theorie von Weyl.4 (5) aus einer voUstandigen Losung W der H a m i l t o n . unter deren Gesichtspunkten sie erst eine unmittelbar verstandliche physikalische Bedeutung gewinnt. Wenn man Ernst macht mit der radikalen Kontinuumsauffassung der Materie. mit der Auflosung des diskontinuierlich abgegrenzten Elektrons in eine stetig in Raum und Zeit veranderliche Feldgrofie. 377 Made (2 a). h. Bd. dafi man namlich in ihr nichts geringeres als einen f o l g e r i c h t i g e n W e g z u r U n d u l a t i o n s m e c h a n i k zu erblicken hat. womit nun allerdings seiner Theorie ihr so pragnanter physikalischer Sinn genommen war und sie dadurch sehr an Uberzeugungskraft verlor. Als „Theorie von de B r o g l i e " bezeichne ich jene noch unvollkommene Vorstufe der Undulationsmechanik. als ihr Urheber bereits wirksam gemacht hat. XLII. Es bedurfte wohl einer ungewohnlich klaren metaphysischen Uberzeugung. wobei die Integrationskonstanten in bekannter Weise so zu bestimmen sind. in welcher die Wellenfunktion der Bewegung e i n e s Elektrons (auf welche wir uns hier beschranken) ^ = e h ' i— 1.J a c o b i s c h e n partiellen Differentialgleichung /dW e '\/dW e \ f hervorgeht. 25 . d. W additivperiodisch wird. K a p i t e l II.2.3. D i e U n d u l a t i o n s m e c h a n i k v o n de B r o g l i e u n d die T h e o r i e v o n W e y l .
wenngleich eine solche Aussage noch in einem sehr problematischen Zusammenhang zu einer ausfuhrbaren Messung stehen mag. was man verlangen kann: die Angabe i r g e n d e i n e s r e a l e n G e g e n s t a n d e s (als „ P r o t o t y p " ) . so hat sich jetzt die Sachlage von Grund auf geandert. von einer solchen kann ja auch in der Elektronentheorie nicht die Rede sein. Aber wenn man i r g e n d e i n e n d e f i n i e r t e n S i n n mit einer metrischen Angabe verbinden will. hatte nicht W e y l in seiner Verallgemeinerung des R i e m a n n s c h e n Raumbegriffs bereits einen Raumtypus geschaflen. auf welche vor allem von Born und seinen Mitarbeitern hingewiesen wurde. ist es unschwer einzusehen. daB. ware vollig hoffnungslos. Ich vertrete durchaus nicht die Auffassung. in die man hier versetzt ist. ein reproduzierbares MaB zu bilden. wenn man untersucht. da man ja gerade in den Elektronen reproduzierbare Maflgrbfien zu besitzen glaubte. ist das mindeste. welche als reproduzierbare Mafistabe die Festlegung einer Mafibestiinmung gestatten konnten. Aber ein solcher realer Gegenstand ist in dem Undulationskontinuum nicht vorhanden. War bisher diese Theorie im Weltbild der diskontinuierlichen Elektronentheorie eine i i b e r f l i i s s i g e Belastung.378 F.lschen RaUmlehre geringes Interesse. keine starren Korper. Denn in diesem schwingenden und fluktuierenden unendlich ausgebreiteten Medium. auf welchen die metrisohe Aussage bereits bezogen ist: Eines Elektronendurchmessers oder abstands usw. Aber da jene Auffassung zunachst jede Interpretation in Raum und Zeit ablehnt..London. sich x ) Es sprechen bekanntlich wichtige Griinde dafiir. lnsofern die Ladungsdichte als eine statistische Gewichtsfunktion umgedeutet wird. hat fiir sie die Beziehung zur W. durch diese de B r o g l i e s c h e und konsequenter durcli die spater zu betrachtende S c h r o d i n g e r s c h e Theorie nahegelegt 1 ) wird. so gelangt man in eine auflerordentliche prinzipielle Scliwierigkeit. Die prinzipielle Lage. daD der ganze Undulationsformalismus statistisch umzudeuten ist. in welchem gerade die Nichtreproduzierbarkeit der Eicheinheit als konsequentes Postulat einer radikalen Nahgeometrie vorgesehen ist. welches geeignet ware. um von Geometrie im atomaren Gebiete zu reden. entstehender und zerfliefiender Wellen nicht anzuwenden. auf die wir hier hinweisen. daB dieselbe Unbestimmtheit hinsichtlich der Anwendbarkeit des Satzes der Identitat. welchen Sinn man iiberhaupt metrischen Aussagen innerhalb des Undulationskontinuums beizulegen hat. scheint mir. . findet man keine unveranderlichen Diskontinuitaten. welches an die Stelle des abgegrenzten Elektrons getreten ist. kein Merkmal im Kontinuum festzuhalten. sich hiniiber iibersetzt. Der Satz der Identitat ist in dem jrwvta get. Man ist geradezu gezwungen. eine a u s f i i h r b a r e Mefivorschrift angegeben werden miisse.ey.
U m0 \o$i e _. Und zwar werde er mit der Stromuhgsgeschwindigkeit der Materie. J ) Diese Auffassung von T. hier sind die dx1 gemafi der durch (7) angegebenen Stromung zu fiihren: 1 « £ { f (5W _ i . \ /air _ e_ * dr c2 . er sei gleich — Also he ini_ j" t_ i l = l0e * i ' 2711 . der Gruppenvierergeschwindigkeit «• = dx* — = ax 1 /dW .T " ) ' . Es sei also jetzt tp = oder iTii i f a i r ( e "• ) (oa) \ 1 a x1 I. 379 auf den allgemeinen Weylschen Raumbegriff zuriickzuziehen und zu versuchen. ihn auf das S c h r o d i n g e r s c h e Kontinuum anzuwenden. wir besaBen bereits einen Mafistab / der sich nach der Weylschen Vorschrift (2 a) verandert. 1926. Da enthiillt sich nun ein einfacher Zusammenhang. Nehmen wir einmal an.Qusintenmechamsche Deutung der Theorie von Weyl. und fiihren ihn im i/>Felde herum. ' (2a) SchlieClich nocli: ich benutze nicht genau das rp aus Gleichung (5).. . Er transformiert sich wie die Eigenzeit. 81. wie es den Vorschlagen von K l e i n . sondern das mit dem Faktor e h versehene fiinfdimensionale tp. = e h Diese Grofie i/> ist zu vergleichen mit dem entlang der Stromung des Kontinuums gefiihrten Weylschen Eichmafie (2a). wobei unter t die Eigenzeit 1 ) zu verstehen ist. d. zuruckgeht.••). Ich behaupte. Hierzu sind noch zwei Prazisierungen zu treffen: In dem Weylschen EichmaC war noch ein Faktor cc unbestimmt 2nie gelassen. steht durchaus in Cbereinstimmung mit der kiirzlich diskutierten Deutung als Winkelkoordinate der Eigenrotationsbewegung des Elektrons (Naturwissenschaften 15. Phys. die auf K u d a r . § 1.\ <PM c / (7) gefiihrt. F o c k und K n d a r entspricht. fur diesen mache ich die Hypothese. Man erhalt: • = i . 15. 1927). ' '.. Denn dieser Drehwinkel ist als eine vom Elektron mitgefiihrte Uhr anzusehen. mit dieser naheliegenden Vorschrift iiber den Weg wird W e y l s S k a l a r I n u m e r i s c h i d e n t i s c h m i t d e m de B r o g l i e s c h e n F e l d s k a l a r ty. 632. f i i e s . Ann.
London.• const j. von a b g e l e n k t e n Lichtstrahlen und Massen oder aber von ihrer g e o d i i t i s c h e n Bewegung in einem R i e m a n n s c h e n Raum zu reden. Differentialgleichung (6) ist der Infolge der H a m i l t o n .. mit einer Lange I vergleichen. wennmansiemitderzugehorigen Strom.380 F. d. Das scbeint ein Widerspruch zu den grundlegendsten Ergebnissen de B r o g l i e s zu sein.und Phasengeschwindigkeit fort. h. sie also erfahrt im elektromagnetischen Felde genau den Einflufi. den de B r o g l i e s c h e n Schwingungsvorgang der Materie und seine Beeinflussung durch die elektrischen Potentiale geometrisch zu deuten durch einen h o m o g e n mit Materie ausgefullten W e y l s c h e n Raum. . der sich so verhalt wie das W e y l s c h e Mafi: d i e k o m p l e x e A m p l i t u d e d e r de B r o g l i e s c h e n W e l l e . scheint mir ebenfalls keine Schwierigkeit zu bieten. welches dispersionsfrei ist und demgemafi fallt hier die Unterscheidung zwischen Gruppen. sondern das runfdimensional erweiterte ^ verwandt.J a c o b i s c h e n Integrand = — mg c2.— — •e h = const. daB wir hier i\>. Man miifite also auch einen konstanten Wert der de BroglieschenWellenfunktion erhalten. Ein weiterer Einwand. Aber das ist hier nicht zutreffend. Man iiberzeugt sich auch leicht unmittelbar. man erhalt: . denn oben wurde nicht genau das d e B r o g liesche. welchen W e y l fiir sein Eichmafi postuliert hat und dem er — als ein leerlaufendes Glied der damaligen Physik — eine metaphysikalische Existenz zuweisen mufite. dessen metrischer Zusammenhang jedoch nicht integrabel ist. i 2 71 i W i —. daB die ebene Welle v> = c h Ui/p \'IP ) (p = —J in der Tat beim Verfolgen mit der Geschwindigkeit v konstante Phase zeigt. Bei fehlendem elektromagnetischen Felde soil nach (2 a) das EichmaC eine Konstante sein. nach welchen die Phasen seiner Wellen mit einer sehr viel groCeren P h a s e n g e s c h w i n d i g k e i t (it. so gibt uns (8) die Moglichkeit. eine Dichte. G r u p p e n g e s c h w i n d i g k e i t (v stets <Cc) verfolgt. Sie also ist' sozusagen das Prototyp des W e y l schen Mafies. Und ahnlich wie es in der Gravitationstheorie in unserem Belieben steht. (8) ' 'o Der physikalische Gegenstand ist gefunden. = —) fortschreiten.
dafi die Erfahrung g e g e n die Nichtintegrabilitat des Eichmafies spricht. so dafi t r o t z d e r N i c h t i n t e g r a b i l i t a t der Streckeniibertragung das Eichmafi an jeder Stelle s t e t s in e i n d e u t i g e r Weise realisiert wird. eine Zusammenfassung von z w e i physikalischen ZustandsgroCen. da in ihr nichts iiber die „Natur" von i! bekannt war. daG jede Strecke als eine komplexe Grofie aufzufassen ist. und man vermutet. Hierdurcli und infolge der Identitat der Wellen . Man hat hierin ein Gegenstiick dafiir zu sehen. dieselbe. § 2. darstellt. durch welche die alte S o m m e r f e l d . wie dem de B r o g l i e s c h e n i/>. wie sich diese Schwierigkeit losen mufi: Die Quantentheorie erlaubt der Materie nur eine diskrete Reihe von Bewegungszustanden.Quantenmechanisclie Deutung der Theorie von Wcyl. dafi diese ausgezeichneten Bewegungen das EickmaB nur derartig zu transportieren gestatten. dafi die Wellenfunktion ip selbst w e s e n t l i c h k o m p l e x aufzufassen ist. das mochte ich noch nicht zur Diskussion bringen. aber infolge der fiinfdimensionalen Erweiterung der Wellenfunktion ist der Schwingungsvorgang dispersionsfrei. Naturgemafier ware es wohl. Aber noch besteht der Einwand. Was es aber nun bedeuten soil. und dafi sich die ganze W e y l s c h e Variabilitat des StreckenmaBes als eine A n d e r u n g e i n z i g d e r P h a s e unter Beibehaltung des Absolutbetrages herausstellt. namlich ibti) und dem Realteil von : In ip. auf den wir oben hinwiesen. Man sieht jetzt bereits voraus. durchgemacht hat. was nur eine Anderung in der Wahl des unbestimmten Faktors u bedeuten wiirde. 381 Man miifite T\> von vornherein nait l~s vergleichen. Innerhalb der Weylschen Theorie konnte eine solche Aussage nicht getroffen werden. und unsere Stromgeschwindigkeit wird infolgedessen identisch mit der Phasengeschwindigkeit. besser gesagt. Es ist hierbei durchaus nicht zulassig. In diesem Sinne ist es auch zu verstehen. sich etwa auf den Realteil zu beschranken. 2 %i dafi im Variationsproblem der Wellenmechanik rjj und ip u n a b h a n g i g voneinander zu variieren sind. dafi der W e y l s c h e n Eichgrofie Z von vornlierein dieselbe Dimension beizulegen ist. In der Tat erinnert man sich an die Resonanzeigenschaft der de B r o g l i e s c h e n Wellen. Eine ernstlichere Schwierigkeit scheint die k o m p l e x e F o r m der Streckeniibertragung dem Verstandnis aufzugeben. daB die Phase bei Riickkehr an den Ausgangspunkt gerade eine ganzzahlige Anzahl von Umlaufen.E p s t e i n s c h e Quantenbedingung von de B r o g l i e zuerst so folgenreich umgedeutet wurde. Diese ist allerdings an die Phasengeschwindigkeit gekniipft. aus dem bier aufgedeckten Zusammenhang zu entnelimen.
benutzt man die bereits in § 1 verwendete Relation: I (U7 *')***=1 m^cu=1 ***ii .{$**• 2 ) Diese SchluBweise ist nicht priizise. Es ist deshalb vielleicht nicht iiberflussig. (9) Um das zu beweisen. funktion ip mit dem Weylschen Mafie erscheint es also bereits erwiesen 1 ). 13. Es wurde auch die Moglichkeit von 2 % i • — ins Auge gefafit. wie sie urspriinglich gemeint war. welchen er spater unter so ganz anderen Gesichtspunkten wieder begegnen sollte. Ich mochte diesen Gegenstand nicht verlassen. a ) E. 1922. ist ein ganzzahliges Multiplum der P l a n c k schen Konstanten: K = £ — 0idxi = nh. so ware man folgerichtig auf das System der diskreten Bewegungszustande der „klassischen" Quantentheorie und ihre de B r o g l i e s c h e n Wellen gefuhrt worden. Phys. als einen Satz der „klassischen" Quantentheorie beweise. f. ohne dafi sie damals in ihrer Bedeutung erkannt wurde.382 F. die uns hier als charakteristischer Satz der Undulationsmechanik entgegentritt. Schon damals also hatte S c h r o d i n g e r die charakteristischen wellenmechanischen Periodizitaten in der Hand. an der Resonanz der de B r o g l i e s c h e n Wellen teil hat und trotz der Nichtintegrabilitat des Differentialausdrucks (2 a) im elektromagnetischen Eelde dennoch zu einer eindeutigen MaBbestimmung an jeder Stelle fiihrt. Hatte man die Eindeutigkeit des Mafibegriffes als eine allgemein anerkannte Erfahrungstatsache der W e y l s c h e n Theorie axiomatisch angeschlossen. 12. von S c h r o d i n g e r 2 ) bereits 1922 als eine „bemerkenswerte Eigenschaft der Quantenbahnen" vermutet und an einer Anzahl von Beispielen demonstriert worden ist. wenn ich es nur entlang der quantentheoretiseh moglichen Materiestromung fiihre. wenn ich diese S c h r o d i n g e r sche Vermutung auch unabhangig von den wellenmechanischen Zusammenhangen. . sie wird sogleich richtiggestellt werden. gefuhrt iiber eine raumlich geschlossene Quantenbahn. aber ihr nicht der Vorzug vor einer nc anderen Wahl von a zuerkannt. London.auch das W e y l s c h e MaB. datl. dafl diese Resonanzeigenschaft des W e y 1 schen Streckenmafles. ZS. ohne darauf aufmerksam zu machen. Es ist also behauptet: Der Streckenexponent des W e y l s c h e n Hafies. S c h r o d i n g e r .
ist dW O X. Es ist 1 Hieraus durch Produktintegration unter Beachtung der Periodizitiitsbedingung: Pi — ist infolge der Bewegungsgleichungcn = — sgleichungcn v jG/uot . also: — c j >  <M<s* = —nh + £ ( . Diese Voraussetzungen (insbesondere die erste) sind offenbar sehr wesentlich und sie werden sich sicher nicht vollig umgehen lassen. . . dxi= — (Ekin f Evot) d t. . Hier verschwinden die Integrale auf der rechten Seite infolge der relativistischen Verallgemeinerung des Virialsatzes*) unter der V o r a u s s e t z u n g . Sie garantieren gewisse stationare Verhaltnisse im Raume. Man sieht aus dieser Ableitung. woraus die Behauptung (9) unmittelbar t'olgt. daC das Potential homogen vom Grade — 1 in den xl ist.Quantenmechamsche Deutung der Theorie von Weyl.dx erhalt man hieraus: l 383 = nh V o r a u s g e s e t z t . dan nur unter z w e i V o r a u s s e t z u n g e n der Eindeutigkeitsbeweis des Weylschen EichmaBes gelingt. die es i i b e r h a u p t e r s t *) Mir ist ein Beweis der relativistischen Verallgemeinerung des Virialsatzes in der Literatur nicht bekannt. Infolge der Quantenbedingungen yt CD ^Jdx1 ^r—. dafi ein Energieintegral existiert. deshalb will ich ihn hier mitteilen. mat erhalt also man diet l Hier verschwindet der Integrand infolge des Eulerschen Satzes liber homogene Tunktionen.«0c2 J/1 (^j + Ekin + E^dt.
welche im allgemeinen vom Bezugssystem durchaus abhangig ist. erscheint aufierordentlich befriedigend. denn ein Transport mit anderer Geschwindigkeit ware quantentheoretisch (bzw. daB man sich mit dem Gewonnenen bereits zufrieden geben kann. g e s t a t t e n . M o g l i c h k e i t fiir d i e A n w e n d u n g d e s S a t z e s dex I d e n t i t a t auf d e n R a u m zu bezeichnen haben. Sie werden daher falsch. Q u a n t e n m e c h a n i s c h e U m d e u t u n s der Theorie von Weyl. daB bei Hnreichend guter Annaherung an die Ausgangspunkte das W e y l s c b e MaB bis auf einen vorgegebenen beliebig kleinen Betrag mit seinem urspriinglichen Wert iibereinstimmt. Ich hoffe. welche nicht meiner Uberzeugung entspricht. Ich habe die. Dann kann man unter geeigneten Stetigkeitsvoraussetzungen beweisen. Die Untersuchungen des vorigen Kapitels erstreckten sich ausdriicklich auf die als „de B r o g l i e s c h e Theorie" gekennzeichnete Vorstufe der Quantenmechanik. Man wird deshalb diese Voraussetzungen als B e d i n g u n g e n d e r . DaB hierbei stets der Transport der Eichstrecke mit der Geschwindigkeit (7) der Materie zu erfolgen hat. so glaube ich doch nicht. wie die W e y l s c h e n Ideen eine nicht vorauszuahnende Verkorperung in den gegenwartigen physikalischen Anschauungen gefunden haben. mechanisch) garnicht moglich. wo beide Theorien auseinandergehen. v o n r a u m l i c h g e s c h l o s s e n e n B a h n e n in d e r Mink o w s k i s c h e n W e l t zu r e d e n . Wenn wir auch gesehen haben. sondern nur^ quasiperiodisch sein. Kontinuumsauffassung der Quantenmechanik hier mit einer E i n s e i t i g k e i t in den Vordergrund gestellt. Man kann aber K a p i t e l III. da hierzu noch wesentlich andere Gesichtspunkte namhaft gemacht werden miissen. Meist werden die Bahnkurven nicht exakt periodisch. In diesem Sinne sind die Ausfiihrungen des folgenden Kapitels durchaus als Provisorium zu betrachten. eine Aussage. Eine nahere Rechtfertigung dieser Zusammenbange und ihren Einbau in eine erkenntnistheoretisch begriindete Theorie des MaBes mochte ich jedoch noch verschieben.384 P. auf den ganzen Zusammenhang unter allgemeineren physikalischen Gesichtspunlvten demnachst zuriickzukommen. wenn man sie unmittelbar auf die S c h r o d i n g e r s c h e Theorie iibertragen wollte — wenigstens in dem Gebiete. zunachst diesen Gedanken mit einiger Konsequenz bis zu Ende zu verfolgen. London. Immerhin schien es mir wiinschenswert. Mehr braucbt man auch nicht zu verlangen. .
dafl wir gegenwartig der S c h r o d i n g e r s c h e n Theorie ihrer Idee nach und wegen ihrer besseren tJbereinstimmung mit der Erfahrung unbedingt vor der de B r o g l i e s c h e n den Vorzug zu gehen haben. zu einem zusammenhangenden Wellenkontinuum Reehnung tragt. denen d e B r o g l i e zunachst nur auflerlich durch (5) eine Welle aufgepragt hatte. In der Wellenoptik dagegen erfahrt ein einzelner Wellenstrahl. . In der geometrischen Optik ist die Betrachtung der einzelnen losgelosten Trajektorien und die der Wellenfronten physikalisch aquivalent. einen gewissen EinfluC durch seine Nachbarn. I n ihrer Diskrepanz mit der Weylschen Theorie haben wir gewiC keinen Mangel der S c h r o d i n g e r s c h e n Theorie zu sehen. daC sie der Tatsache der „Eingemeindung" der Trajektorien der klassischen Mechanik. Wenn man beachtet. 385 jedenfalls bereits sagen. so kann kein Zweifel sein. Bei Zerlegung in imaginaren und reellen Bestandteil lautet die S c h r o d i n g e r s c h e Wellengleichung fiir lHlw TP = \tp\e h (T'Freell): / h \ 3 DI f I ' d f. Es ist keine Frage. wenn sie die Wellenfunktion i\t anstatt durch eine J a c o b i s c h e Differentialgleichung (6) durch eine Wellengleichung beschreibt. Man kann den Fortschritt zur S c h r o d i n g e r s c h e n Form der Wellenmechanik dahin charakterisieren. dafi sich die Abweichungen charakteristisch bei kleinen Quantenzahlen einstellen.. . wenn er einer Front von Strahlen e i n v e r l e i b t wird. da beide Theorien dort ineinander ubergehen. Die zweite Gleichung ist die Kontinuit'atsgleichung des Stromes. '• Zugleich wird hier auch 11 * sichtbar. dafl es sich um ein Problem mit zwei unbekannten reellen Funktionen handelt. ist die charakteristische Aussage der S c h r o d i n g e r s c h e n Theorie. dafl unsere Resultate a s y m p t o t i s c h richtig bleiben miissen in der Grenze grofier Quantenzahlen. dessen vier Komponenten durch die geschweiften Klammern eingefafit werden. worauf die Schwierigkeit zuruckzufiihren sein wird: Die W e y l s c h e Theorie ist . c /dW (d W e \ (d W ' e ^ \] e \ ' (10) In dieser Darstellung erkennt man den Gegensatz zur de B r o g l i e s c h e n 0\ip\ Theorie in dem Auftreten des Gliedes .Quantenmechanische Deutung der Theorie voa Weyl. DaC dieser EinfluS zum Ausdruck kommt.
Madelung. Ann. Bisher war angenommen. 1927. Es ist also zu erwarten. . in der elektronentheoretischen Auffassung hinter undurchdringliche Grenzflachen aus dem Felde verbannt oder in die Singularit'aten desselben verwiesen. 265. Naturwissenschaften 15. daB die vier Potentiale 0it welche eine vollstandige Beschreibung des elektromagnetischen Feldes liefern. 322. d.<'('V^5^iJ3 (> » !) E. sie muB ihrerseits entsprechend der quantenmechanischenKorrektur der klassischen Gesetze modifiziert werden. Die Materie. Es ist demzufolge von ihr gar nicbt zu erwarten oder zu verlangen. an der jetzt veralteten W e y l s c h e n Theorie den entsprechenden Schritt zu vollziehen. dafl sie auf die S c h r o d i n g e r s c h e Theorie bereits pafit. welches in vieler Hinsicht — vor allem in der Darstellang durch ein Variationsproblem') — symmetrisch den FeldgroBen c£>. E h r e n f e s t . S c h r o d i n g e r . gegenlibersteht.386 F. f. in welcher Richtung die Korrektur des Weylschen MaBes geschehen wird. einzig fiir die Streckenverschiebung maBgebend seien (2 a). Jetzt hat sich die Sachlage insofern geandert. zu beriicksichtigen sein wird. ihrer ganzen Kompetenz naoh sozusagen auf die klassische Mechanik und somit audi auf die ihr zugeordnete de B r o g l i e s c h e Theorie zugeschnitten. 161. Phys. und wahrend man in der Weylschen Theorie sich mit Recht einen Mafistab im „leeren" Raum nur von den dort herrschenden elektromagnetischen Potentialen beeinflufit dachte. M a d e l u n g 3 ) hat das „Potential" dieser inneren Wirkung des i/>Feldes auf sich selbst angegeben. Ich mochte als relativistische Verallgemeinerung desselben vorschlagen: ^=»'.) 3 ) E. welcher von de B r o g l i e zu S c h r o d i n g e r fiihrt. 40. Phys. Man kann voraussehen. wird jetzt dem Umstand Rechnung zu tragen sein. ist jetzt iiber den ganzen Eaum ausgebreitet. Superpositionsprinzip! Dennoch scheint die Eigenschaft der Undurchdringlichkeit in Form des Pauliverbots ihren quantenmechanischen Ausdruck zu finden. London. 82. ZS. (P. 1926. Die Aufgabe muB vielmehr sein. als zu den vier ZustandsgrbBen des Feldes 3>j als fiinfte das S o h r o d i n g e r s c h e tjj getreten ist. daB die alte Trennung zwischen der „ u n d u r c h d r i n g l i c h e n " Materie und dem xsvbv aufgehoben ist und man sich s t e t s sozusagen im Innern der alles durchdringenden 2 ) neuen Substanz  tp  befindet. ) Denn y geniigt einer linearen Differentialgleichung. 1927. die allein vori  rp  abhangt. daB aufler den auBeren elektromagnetischen FeldgroBen noch eine innere.
dx^ ist zwar eine r e l a t i v i s t i s c h e Invariante. 70e * J 2 j o ^ d I ( i3) Verallgemeinerung des Weylschen Strecken Urn die Identitat von (13) mit der S c h r o d i n g e r s c h e n Wellenfunktion nachzuweisen. . langs welchen "Weges das verallgemeinerte StreckenmaB (13) zu fiihren ist. welche nicht unabhangig von den iibrigen dxt ist. aber keine Eichinvariante (TJbergang zu einer anderen Eicheinheit andert dr). 387 Das W o r t „ Potential" ist mit Vorsicht zu gebrauohen. — Offenbar sind in diesem Sinne die funfdimensionalen Ansatze von Kaluza zu verstehen. <50 entspricbt nicht etwa dem „skalaren" Potential <E>4. daC / = Jie quantenmechanische maGes darstellt.Quantenmechanische Deutung der Theorie von Weyl. Das AVeltlinienelement di bzw. Wenn man iiberhaupt einen EinfluC auf das Eichmafi annehmen will. welches relativistiscb als z e i t l i c h e K o m p o n e n t e eines Vierervektors figuriert.~T~~ — — d% dx c (12 ) ') Das Auftreten dieser funfdimensionalen quadratischen Form ist im Sinne der Weylschen Forderung der Eichinvarianz ganz konsequent. so wird man vermuten. Hierbei ist aber zu beachten. Man wird wieder den Transport mit der Stromungsgeschwindigkeit der Materie vorschreiben wollen. Die derart abgetrennten Geschwindigkeitskomponenten wiirden namlich wegen (10j) nicht die Identitat der Vierergeschwindigkeit 2 ) M M = * dxkdxk T. obwohl die Darstellung des Stromes in der zweiten Gleichung (10) die Abtrennung des Faktors etpip als Ruhladungsdichte nahe legt. 2 ) Wenn nichts anderes angegeben. n i c h t v o n i h r e r R i c h t u n g . sondern sich ihnen durch die Bedingung dx* f dx* f dxs2 + dx* f dxb2 = 0 (12) gleichberechtigt zugesellt*). wohl aber ist das Verschwinden der quadratischen Form (12) eichinvariant. so kann er n u r v o m B e t r a g e der vierdimensionalen Streckenverschiebung abhangen. sondern ist auch relativistisch ein invarianter S k a l a r . Fiihrt man dementsprechend durch das Weltlinienelement dxb = cdx (r = Eigenzeit) eine flinfte Koordinate ein. sind im folgenden die Suraraationen iiber gleiche Indizes stets von 1 bis 4 wie bisher zu verstehen. Dementsprechend kann <P6 auch nicht die Streckenanderung langs einer bestimmten Weltrichtung regieren. daG jetzt die Komponenten ul der Vierergeschwindigkeit nicht durch (7) gegeben werden. miissen wir zunachst angeben.
(7a) wobei der Faktor f \2m/ m02c2 \if>\ \ m0c2 y als „ Ruhladungsdichte" abgetrennt ist. d. + ( * s ) dx5 (11 a) ergibt:   ! £ f s ! J. <P2. <P3. (5a)]. 2R5. = L*J 'o ' Letzteres wegen (10 a). dail <I>6 seinerseits noch selbst eine erst zu bestimmende Unbekannte ist. sondern Man erhalt also zuniichst nicht tyjl = y = y. P. 1927. . Bekanntlich ist es ein noch unverstandenes Wunder. — I ** I I . London. In dieser Bezeichnung erhalt man m0c2 ( l M (11a) \ e^W und die erste S c h r o d i n g e r s c h e Gleichung lautet in fiinfdimensionaler Fassung J ) : e<X>b = sN /dW e \ /dW e \ Wir vergleichen jetzt die Strecke I (13) entlang der StrSmung (7 a) mit dem S c h r 6 d i n g e r s c h e n Skalar ip. Phys. dx. * / 1 dxr. Man erhalt fiir xp/l (7 a) ergibt: 10 —7— >. •wie man erwarten mfifltc. 82. Ann. Es ist vielmehr zu schreiben dxk _ uk i i>4> — e fdW \ 7 e ®*  . konst. Schrbdinger. warum das gleiche nicht fiir die Potentiale <Pt. (E. *«\ /aw. $ 4 gilt.) s— ist = m 0 c [vgl.388 erfiillen. (8 a) *) Hierbei ist zu beachten. _ i „ \ . £ ( » Z _ i.
. fiihre ich statt ihrer <X>* = ®k — K—r. Hierbei ist die 5. d. S t u t t g a r t . Hochschule. so folgt i[)/l = konst. Physik. 389 welches eine eindeutige Ortsfunktion ist 1 ).•.?. 6 = o c. techn. c ll. In \ip\ als Potentiale ein. \ . .' = 0 1. Komponente des Fiinferstroms . 1. <PA ist parallel dem Fiinferstrom j i = — y y I . so dafi wir den tlberlegungen des 2.Quantenmechanische Deutung der Theorie von Weyl. Ich verdanke diese schiine Forimilierung einer Mitteilung von Herrn A. c c \ixl I ' m \ixl / dxl soil parallel dem Fiinferstrom j l gewahlt werden. Inst. 27: Februar 1927. ') Man kann diese Beweisftihrung im Sinne der fiinfdimensionalen Geometrie sinngemafier folgendermafien aussprechen: ZbW c \ e — /iW e \ I . Der Fiinferstrom ist orthogonal auf sich selbst I ^>] .JT. also ist j( auch orthogonal auf dx » /3 W und also >_: I—. Kapitels hier nichts hinzuzufugen haben. Aber die Potentiale <2>t sind nur bis auf einen additiven Gradienten physikalisch festgelegt. 'Adx = 0. <P. Die auf der Resonanz der Wellen beruhende Eindeutigkeit des mit der Stromung mitgefiihrten EichmaBes iibertragt sich natiirlich jetzt ohne weiteres aus der de B r o g l i e s c h e n auf die S c h r o d i n g e r s c h e Theorie. was die elektromagnetischen Feldstarken unberiihrt laBt. Lande.
II.and rightly so . Chap. characteristics of the scaling relations of space . also a preliminary summary in Naturwissen. L o n d o n in Stuttgart (Received on 25 February 1927. Or. is a "rigid" scale.& at a point. it had to be assumed that the scale which is used at each point to determine the coefficients gik of the metric fundamental form. but not their absolute values. Chap. In order to make sense of this statement by Riemann. unless the quantities Iik ~ dx* dx' [6) Presented partly at the meeting of the Gauverein Wiirttemberg of the German Physics Society Stuttgart. 15. In general.1 Quantum mechanical interpretation of Weyl's theory 1 By F . (1) where the proportionality factors <pi are functions of the position in space. the gauge scale is pathdependent (nonintegrable). and that one can only determine sensibly the ratios of the </. ds2 — gikdx'dxk. Weyl's Theory As is generally known. i. I: Weyl's theory. In contrast to this Weyl claims . § 1. Chapter I. Thus. has been completed by Weyl in a remarkably beautiful and simple way. first conceived by Riemann. 1927 .) Chap.that the assumption of such a rigid scale is against a purely local geometry. he makes the ansatz for the change dl of a length scale I under an infinitesimal shift dx{: dl = lipidx'. on 18 December 1926. if one integrates (1): l = l0eS^dx' (2) (/0 = / at the beginning of the shift). Nonintegrability does not exclude uniqueness.e. 187. cf. Riemann's notion of space can be viewed as the lifting of the prejudice that the curvature conditions at one point of space control the curvature at all others. III. recently the idea of a "purely local geometry". The identity of V' and Weyl's gauge distance.similar to the gij. Broglie's wave mechanics and Weyl's theory. § 2. Q u a n t u m mechanical reinterpretation of Weyl's theory.
4 ox 1 (4) ox % ox" The formal match of these four equations with the system of Maxwell's equations rot € + U = 0. According to their definition (3) the quantities /. which prompted Einstein to his geometric interpretation. a remarkably clear metaphysical conviction was needed. I do not need to dwell on this abstract development of the theory. and only in the light of this does it gain a directly comprehensible physical meaning. no such fact was known: there was no reason to think of a universal influence of the electromagnetic field on the socalled rigid scales (or clocks). however. namely./. removed the theory's striking physical meaning. obey the identity (assume that the manifold is of dimension four): 7rr + 7rr + § l = 0> *"#M'. He held on to his belief and avoided a discussion of the aforementioned contradictions by a somewhat obscure reinterpretation of the notion "real measurement". c div U = 0.2.3. which Weyl assumes in a magnetic field. Thus. I will rather show that exactly this concise original version of Weyl's theory contains even more possibilities than has been used by its author. the principle of equivalence of inertial and gravitational mass. with the components $ . By logically extending the geometric interpretation of gravitation as variable curvature in Riemannian space. On the contrary.in defiance of such elementary experiences . and also other formal analogies have led Weyl to the conclusion that the ipi should be identified. u p to a constant proportionality factor. one has to write: / = l0ea J 0* x' (a = proportionality factor). with the electromagnetic field strengths £. characterized by the variability of the gauge scale. *\M = 1. and accordingly the fa. also as a property of the scaling relations of space. Apparently. One must admire the tremendous courage with which Weyl tracked down his theory of the gaugegeometric interpretation of electromagnetism solely on the grounds of this completely formal relation: in the case of the theory of gravity it was a physical fact. contradicting to the nonintegrable measure (2a). In the theory of electricity. namely that one has to view it as not less than the logical path to wave mechanics.not to deviate from the thought that Nature must use the geometric possibilities given to her. and it became far less convincing. however. of the electromagnetic fourpotential. . for example atomic clocks do represent scales.H. whose independence of the history has been demonstrated by the sharpness of the spectral lines. Weyl imagined the remaining type of physical forces. that caused Weyl .2 vanish. the electromagnetic field. This.
W is additively periodic. as it is suggested by de Broglie's theory and more consistently by Schrodinger's theory2 to be discussed below.e. which took the place of the discrete electron. since this point of view initially rejects any interpretation in space and time. The theorem of identity is not to be applied in the irdfTa get of forming and dissipating waves. reappears. infinitely extended medium.2. that the whole wave formalism should be reinterpreted statistically. But if one wants to assign a defined meaning to a metric specification. fundamental difficulty if one analyzes which meaning should be attached to metric statements within the wave continuum. The basic situation one finds oneself in would be completely hopeless. which we point out here.R. no rigid bodies which. which were pointed out primarily by Born and his collaborators. although such a statement might still have a problematic relation to a feasible measurement. could allow a determination of a measure. in the continuum no attribute is to be fixed that would qualify as a reproducible measure. and also the dissolution of the discrete electron into a field variable that depends continuously on space and time. There the wave function describing the motion of one electron (to which we will restrict ourselves here) ^=eT f f W 1=1. i.3. Insofar as the charge density is reinterpreted as a statistical weight function it. there are no invariable discontinuities. had Weyl not already in his generalisation of Riemann's notion of space created a type of space in which exactly the impossibilty to reproduce the gauge unit is provided as a consistent postulate of a radical local geometry. etc.4 (5) is determined by a complete solution W of the HamiltonJacobi partial differential equation fdW Wc )\M^) ^ e^ \ fdW 9i e A = 22 . However. For in this oscillating and fluctuating.. which the metric statement is already related to: an electron diameter or distance. where the period is an integer multiple of Planck's constant. . then one is faced with an extraordinary. While until now this theory was an unnecessary burden in the world view of the discontinuous It is known that there are important reasons. after all one cannot say that such a prescription exists in the electron theory. as reproducible scales. Yet such a real object is not present in the wave continuum. it has little interest in the relation to Weyl's theory of space. (6) where the integration constants are to be determined in a known fashion such that i\) is a singlevalued function of space. is not difficult to see that the same uncertainty with respect to the applicability of the identity theorem. If one takes the radical continuum hypothesis seriously. I by no means take the view that a practicable measurement must be given in order to speak of the geometry in the atomic region.3 Chapter II: De Broglie's wave mechanics and Weyl's theory I will call the still imperfect precursor of wave mechanics "de Broglie's theory" . it seems to me that the least one can require is the specification of some real object (as a "prototype").
1927). multiplied by a factor e h m"° T . One gets I k ' here the displacements dx1 are to be given by the flow (7): = I ^{J(f5J*. 8 1 . corresponding to the proposals by Klein. it is moved with the flow velocity of matter. Then a simple relationship is revealed. and we move it around in the rp field. (2a) Also: I do not use exactly the i[.> of equation (5). and one gets: ^ = Ie2^. where r is to be understood as the proper time. 3 Thus let now ^. 15. It transforms like the proper time. is indeed in agreement with the recently discussed interpretation as an angular coordinate of the selfrotational motion of the electron (Naturwissenschaften.)(l^t*')^+ m ^} lo" As a consequence of the HamiltonJacobi differential equation (6) the integrand is = — raoc2. For this rotation angle is to be viewed as a clock carried by the electron. Ann. two issues have to be made more precise: In Weyl's measure the factor a was left undetermined. the group velocity dr' dr 1 /BW mo v dxi e c \ / I claim that with this quite natural prescription for the path Weyl's scalar / becomes numerically identical with de Broglie's field scalar i/>. 15.4 electron theory. 632. 1926.const = c o n s t (8) 3 This sense of T. I = l0e'^ I ^'d1'. which was believed to provide exactly these reproducible scales. Fock and Kudar. Namely. which goes back to Kudar. § 1. Let us assume we already had a scale / which changes according to Weyl 's prescription (2a). One is almost forced to withdraw to Weyl's general notion of space and to try to apply it to Schrodinger's continuum. d. .Thus. but the fivedimensional ip. F ° r this. for this I propose the hypothesis that it be equal to '^r. Phys. = or f 2f(H'+m„c 2 r) ' ^ This quantity has to be compared with Weyl's gauge scale (2a) which is carried along the flow of the continuum. the situation has fundamentally changed now.
is not integrable. I do not want to discuss as yet. One should compare ip with /~ 3 from the outset. But this is not the case here. is found: the complex amplitude of the de Broglie wave. And as in the theory of gravity we are free to speak either of deflected light rays and masses. It is indeed not possible to restrict oneself to the real part. and that the whole of Weyl's variability of the distance measure turns out to be a change solely of the phase at constant modulus. however. namely that here we compare ip. or rather.5 The physical object. namely VV" a n d the real part of ^ r l n ^ .had to assign a metaphysical existence. § 2. group velocity (v always < c). but the fivedimensional extended ip was used above. It is so to speak the prototype of Weyl's scale. One can already see how this difficulty needs to be resolved: quantum theory allows matter to have only a discrete series of states of motion. But there is still the objection we pointed out above. mnt' 2 \ <P = e " V v 7 ^ V^ j (/?=) indeed has a constant phase along a trajectory with velocity v. whose metric connection. which would only imply a change in the choice of the undetermined factor a. That in the variational problem of wave mechanics rjj and tp are to be varied independently is also to be understood in this way. it is the wave that experiences exactly the influence that Weyl has postulated for his gauge scale and to which he . so does (8) allow us to interpret the oscillation of matter and the influence exerted upon it by the electric potentials geometrically as a homogeneous Weyl space filled with matter. This seems to be in contradiction with the fundamental results of de Broglie. Therefore. A more serious difficulty for the understanding seems to be the complex form of the path connection. which behaves as Weyl's measure. it represents a > collection of two physical quantities. one should also get a constant value for de Broglie's wave function if one follows it with the associated flow velocity. or of their geodesic motion in a Riemann space. Within Weyl's theory one could not make such a statement. with no distinction between group and phase velocity. i. because there one did not know anything about the "nature" of I.as a redundant element of the physics at that time .e. in the absence of an electromagnetic field the gauge scale should be a constant. One can easily convince oneself directly that the plane wave 2 TT i / "* Oc J. with a length / does not seem to present any difficulty to me. According to (2a). and this is dispersionfree. and one suspects that these distinguished . a density. It would probably be more natural to learn from the connection revealed here that the same dimension has to be assigned to Weyl's gauge scale / as to de Broglie's ip right from the outset. that experience speaks against the nonintegrability of the gauge scale. since not exactly de Broglie's. according to which the phases of his waves propagate with a much higher phase velocity (u — c2/v). Another objection. Yet what it means that any spatial interval has to be viewed as a complex quantity. One has to view this as the counterpart of the fact that the wave function V itself is to be understood as intrinsically complex.
13. Because of this and the identity of the wave function W with Weyl's measure. Schrodinger had appreciated the characteristic wave mechanical periodicities. as a theorem in "classical" wave mechanics. without its relevance being recognized. was conjectured by Schrodinger5 already in 1922 as a "remarkable property of quantum orbits". and demonstrated in a number of examples. one is reminded of the resonance property of de Broglie waves. f. yet. Schrodinger. so that in spite of the nonintegrability of the connection at every point the gauge scale is unambiguous. is linked with the phase velocity. it seems already proven4 that also the Weyl measure. that by which the old SommerfeldEpstein quantization condition was first reinterpreted so influentially by de Broglie. 1922. and despite the nonintegrability of the differential expression (2a) it leads to an unambiguous assignment of a measure at every point. integrated along a spatially closed quantum orbit is an integer multiple of Planck's constant: I ^idxi = nh. .was also considered. E . It is therefore probably not unnecessary for me to prove this conjecture of Schrodinger's as it was originally intended. as a consequence of the fivedimensional extension of the wave function the oscillation is dispersionfree and therefore the flow velocity becomes identical to the phase velocity. Phys. In order to prove this. which he would meet again later from quite different points of view. however. Indeed. This. 12. but its advantages over other choices was not acknowledged. shares the resonance of de Broglie waves. I don't want to leave this matter without pointing out that the resonance property of Weyl's measure for distances. If the uniqueness property had been added axiomatically as a generally accepted empirical fact. Already at that time. then one would have logically been led to the system of discrete states of the "classical" quantum theory and its de Broglie waves. ZS. which we encounter here as a characteristic feature of wave mechanics. one employs the relation already used in §1: (9) As a consequence of the quantum conditions E J® ——^dx* = nh dx' 4 5 This method of inference is not precise. if I move it only along a matter flow that is quantum theoretically possible. The claim is: the exponent of the Weyl measure. The possibility of a — l^ij. independently of the wave mechanical context. it will be corrected immediately.6 motions allow for the measure to be transported only in such a way that on return to the starting point the phase has completed an integer number of rotations.
Then one can prove under appropriate continuity assumptions that for sufficiently good 6 No proof of the relativistic generalisation of the virial theorem in the literature is known to me.1 in the x%. only these allow us to speak of spatially closed orbits in the Minkowskian world in the first place. They guarantee certain stationary configurations in space. one is forced to call these assumptions conditions for the possibility of the application of the identity theorem to the space. so one obtains " + Epot ) dt Here the integrand vanishes as a consequence of Euler's theorem on homogeneous functions. that the proof of the uniqueness of the Weyl measure succeeds only under two assumptions. It is /H^T^*)/^r f(pdi^ Because of the equations of motion £*• = + #pot I dt + Epot dt From this by integration by parts and with the periodicity condition in mind: + #pot I p < dt ° . m 2 °° v . . these assumptions (in particular the first) are rather essential. Here the integrals on the RHS vanish due to the relativistic generalization of the virial theorem6 if it assumed that the potential is homogeneous of degree . which in general depends on the frame of reference.7 one obtains from this i i^idxA . 1 .{^j + £" kin + £ p o t dt. and it is not possible to avoid them completely.(Ekin thus: * Qidx* = nh + I I m0c2J + Epot)dt. so I would like to communicate it here.& d t dW — dxA . One can see from this derivation. Usually the orbits will not be exactly periodic. a statement.~c*idxi)=nh=f Assuming that an energy integral exists. from which the claim (9) follows immediately. Apparently. but only quasiperiodic. Thus.
It seems extraordinarily satisfying that the transport of the gauge distance always has to happen at the velocity (7) of matter.at least where the two theories diverge. However. Chapter III.At the same time it becomes clear that this is a problem with two unknown real functions. since for this other quite different issues need to be addressed.)(i^f*0 +m2c2=o ' &{l«Ki£!«*)} = 0 (10) In this representation one recognizes the difference to the Broglie theory: the appearance of the term rrr. which de Broglie had imposed upon a wave only superficially by (5). I would like to postpone a closer justification of these connections and their implementation into an epistemologically founded theory of the measure. One could characterize the progress towards Schrodinger's form of wave mechanics by saying that this form takes into account the "incorporation" of the trajectories of classical mechanics. More than this is not needed anyway. Q u a n t u m mechanical reinterpretation of Weyl's theory The analysis of the previous chapter was explicitly restricted to the precursor of q u a n t u m mechanics. I hope to return to the whole context from a more general physical point of view in the near future. one can at least say that our results have to be asymptotically correct in the limit of large q u a n t u m number. it seemed desirable to me first to follow up this thought carefully to the end. however. a single ray experiences the influence of its neighbours when it is merged into a wave front. The second equation is the continuity equation for the current. Although we have seen how Weyl's ideas have been incorporated into current physical conceptions. It would thus be wrong. when it describes the wave function ip by a wave equation rather than the Jacobi differential equation (6). I do not believe that one can already be satisfied with what has been gained. The expression of this influence. if one were to apply it directly to Schrodinger's theory . whose four components are given in the curly brackets. I have emphasized the continuum interpretation of quantum mechanics in a onesided way. into a connected wave continuum. In geometric optics the consideration of the individual decoupled wave is physically equivalent to the consideration of the wave fronts. is the characteristic statement of Schrodinger's theory. Nevertheless. However.'. . In wave optics. In this sense the remarks in the next chapter have to be regarded as being provisional. After decomposition into imaginary and real parts Schrodinger's wave equation for ip — \ip\e h {W real) reads: (^) 2 lf l +(^f*. which does not correspond to own my belief. because there both theories coincide. which could not have been predicted. which I termed the "de Broglie's theory .8 approximations to the initial starting points the Weyl measure agrees with its original value up to a given. arbitrarily small amount. for a transport with a different velocity would q u a n t u m theoretically (or mechanically) not be possible.
Phys. I would like to propose as a relativistic generalisation of it: One needs to be careful in using the word "potential". it can be expected that in addition to the external electromagnetic field variables. The task should rather be to change the now outdated Weyl theory analagously to the step from de Broglie to Schrodinger. then there can be no doubt about what the difficulty reduces to: Weyl's theory is in all its competence so to speak tailored to classical mechanics and therefore also to de Broglie's theory associated to it.. 82.9 There is no question that we currently really have to prefer Schrodinger's theory to de Broglie's due to its basic principle and because of its better match with experience.• there is also the fifth. Madelung. One can foresee in which direction the correction of Weyl's measure will happen. which relativistically plays the role of the time component of the fourvector. $5 does not correspond to the "scalar" potential $4. Weyl's theory must be modified according to the quantum mechanical corrections of the classical laws. If a fifth coordinate is introduced accordingly by the world line element dx$ = cdr E . Madelung9 has given the "potential" for this internal action of the ^>field. 322. not on its direction. Matter. 1926. which in the electrontheoretic approach is banished from the field behind impenetrable boundaries. it can only depend on the modulus of the fourdimensional displacement. it cannot be expected or required of it that it is already compatible with Schrodinger's theory. which provide a complete description of the electromagnetic field. $5 cannot determine the change of the distance along a certain world direction. f. Now the state of affairs has changed insofar as in addition to the four state variables of the field $. Phys. the property of impenetrability seems to find its quantum mechanical expression in the form of Pauli's exclusion principle. we do not have to view the discrepancy with respect to Weyl's theory as a flaw of Schrodinger's theory.is on a footing with the field variables <J>S. or removed into its singularities. ZS. 40. 265. d. 7 . 9 E . Surely. now it has to be taken into account that the old distinction between the "impenetrable" matter and the XEvbv does not exists anymore and that one is always so to speak in the interior of the new substance \i\>\ that can penetrate 8 everything. and rightly so. Superposition principle! Nevertheless. 1927 For 4' obeys a linear differential equation. If one wants to assume any influence on the displacement at all. but it is a relativistic scalar: accordingly. and while in Weyl's theory one thought of a measure in "empty" space being influenced by the electromagnetic potentials. Schodinger's ip. which only depends on \fy\. Schrodinger. Thus. If one takes into account that the deviations appear typically at small quantum numbers. Thus. are responsible for the shift in distance (2a). Ann. is now distributed across the whole space. which in many respects . Up until now it was assumed that only the four potentials $.. one also needs to take into account an internal one.in particular in the representation by a variational problem 7 .
$ 2 .**' I (13) is the q u a n t u m mechanical generalisation of Weyl's measure for distances. d. one would like to prescribe the transport with the flow velocity of matter. Ann. T h a t is to say. 12 Here it has to be noted that $5 itself is a variable. it is not a gauge invariant (transition to a different gauge unit changes dr). 82. but the vanishing of the quadratic form (12) is gauge invariant. (E. Although the world line element dr. Phys. due to (10i) not obey the the fourvelocity identity u t ukuk dxk dx n = ~—— = c2 dr dr je_ p /i r>/\ 12' One should rather write dxk _Uk__ihp_ dx$ c fdW_ _ e c \ J TTIQC \dxk where the factor =e^(l_Lj.8j is separated as "charge density at rest" In this notation one obtains e $ 5 = m0c2 1.10 ( r = proper time). In order to prove the identity of (13) with Schrodinger's wave function we first have to specify along which path the generalised distance measure has to be transported. is a relativistic invariant. 11 Unless stated otherwise in the following the summation over same indices are always understood to run from 1 to 4. or dxs. $ 3 .)   ^ is = m0c [cf. which is not independent of the other dxt. that the components u' of the fourvelocity are not given by (7). Again. still to be determined. 265. it has to be noted. However. $4. the velocity components thus defined would.c J ( a ? . As known. by the relation dx\ + dx\ + dx\ + dx\ + dx\ = 0 then one will conjecture that 5 (12) I = lo exp x/£. (5a)] .**) = ° (10«) 10 The appearance of this fivedimensional quadratic form is entirely consistent with Weyl's claim for gauge invariance. in contrast to what one would expect. Apparently the fivedimensional ansatze by Kaluza have to be understood in this sense. it is a miracle still not understood why this is not true for the potentials $ 1 . 1927. but is on an equal footing with them 1 0 . Schrodinger. even though the representation of the current in the second equation (10) suggests the separation of the factor eipip as charge density at rest.£= V eipip/ (14) (Ha) and the first Schrodinger equation in fivedimensional form 1 2 reads ^ fdW e A fdW r e^\ £ 1^7 .
(dW_ _ e A f dW_ __ e_ \ h J p mc tr* \dxi c J ydrS c 'J /<W _ e \ dd5 c = I " exp Jo ' The last is due to (10a).11 We now compare the distance / (13) along the flow (7a) with the Schrodinger scalar %j>. I owe this nice formulation to a communication by Mr A. are physically determined only up to an additive gradient.^ /? mc V dx{ c J \ dxl c. initially one does not obtain ijj/l = const. is now carried over naturally from the de Broglie's to Schrodinger's theory. J (11a) gives \m_ J in^_e_y. which leaves the electromagnetic field strengths unaltered. = ** he d lnV 2me dxk as potentials. (ta) which is an unambiguous function depending on the position. then it follows 4>/l = const. .. J. the potentials <I>. Thus. Lande. 13 One can formulate the proof in the sense of fivedimensional geometry more accurately as follows: —f — <£>. but * = M. Dept. Stuttgart. Thu. 1 is parallel to the fivecurrent ji = ^ip1!1 ( dx'£ * £ c $ t V dx' is to be chos. Phys. f^e_ exp ~~h 7 . Here the 5th component of the fivecurrent is j'5 = QC. 27 February 1927. J dx' — 0.s. The singlevaluedness of the measure that is carried with the current and which is based on the resonance of the waves. The fivecurrent is orthogonal to itself I J j is also orthogonal to dx' and thus VV I j ~ — <J>. parallel to the fivecurrent. Univ. Techn. For ip/l one obtains ±=\±\ • e x p I I ~h 1 V V oxi c (7a) gi (W_ _ e A / c W _ e \ 2mi t y . so that we do not have to add anything to the considerations of chapter II. 13 However. if I introduce instead * . 'I' jjj' = 0 ) .
) Auf den folgendea Seiten mochte ich auf einen einfachen Zusammenhang hinweisen zwischen der von Kaluza 1 ) vorgeschlagenen Theorie fur den Zusammenhang zwischen Elektromagnetismus und Gravitation einerseits und der von de Broglie 2 ) und Schrodinger 3 ) angegebenen Methode zur Behandlung der Quantenprobleme andererseits. d. Phys. eine kurze Darstellung von der funfdimensionalen Relativitatstheorie zu geben. x4. 966. fange damit an. 3 ) E. de Broglie. Wir schreiben dasselbe: A 6 — V'Syikdxidxk. so kommt man eben zu den obenenvahnten quantentheoretischen ilethoden. wie uberall im folgenden. Die 15 GroBen yik sind die kovarianten Komponenten eines fiinfdimensionalen symmetrischen Tensors. (Eingegangen am 28. Hierbei bezeichnen x° . Bd. die fiinf Koordinaten des Raumes. d.. Von Oskar Klein in Kopenhagen. indem die Materie als eine Art "Wellenausbreitung betrachtet wird. Die Bewegungsgleichungen der elektrischen Teilchen nehmen hierbei auch in elektromagnetischen Feldern die Gestalt von Gleichungen geodatischer Linien an. Sitzungsber.k eines Linienelementes von einem Riemannschen Raum. 1921.der gewohnlichen Relativitatstheorie zu kommen. x3. Akad. eine Summation iiber die doppelt vorkommenden Indizes von 0 bis 4 angibt. die als eine Verallgemeinerung der gewohnlichen "Wellengleichung angesehen werden kann. 79. Kaluza. x2. stets den gewohnlichen Zeitraum charakterisieren. Die Theorie von Kaluza geht darauf hinaus. Theses. April 1926. Urn von denselben zu den GroBen gik und <p. fur welches wir einen vom Koordinatensystem unabhiingigen Sinn postulieren. 59 2 Zeitschrift fur Physik. (1) wo das Zeichen Zl. 22. Zweitens diirfen die GroBen i) Th.. Berl. die zehn Einsteinschen Gravitationspotentiale gik und die vier elektromagnetischen Potentiale fi in Zusammenhang zu bringen mit den Koeffizienten y. Ann. S. 361 und 489. 1925. der anBer den vier gewohnlichen Dimensionen noch eine fvinfte Dimension enthalt. Wenn dieselben als Strahlengleichnngen igedeutet werden. Schrodinger.. . Ich. (10) 3. aber in einigen Punkten von derselben abweicht. XXXVII. Phys. sagen wir x1. Paris 1924. d. bei denen die fiinfte Dimension rein harmonisch auftritt mit einer bestimmten mit der Planckschen Konstante zusammenhangenden Periode. Werden nun solche Losungen dieser Gleichung betrachtet. kommt man fast von selbst zu einer partiellen Differentialgleichung zweiter Ordnung. 1926. Funfdimensionale Relativitatstheorie. Betrachten wir ein fiinfdimensionales R i e m a n n s c h e s Linienelement.895 Quantentheorie und funfdimensionale Relativitatstheorie. die sich nahe an die Theorie von K a l u z a anschlieBt. xi. Ann. Erstens miissen vier der Koordinaten. ) L. miissen wir gewisse spezielle Annahmen machen. § 1.
0' anstatt x°' schreiben sollen.= Const. wie ublich. namlich ') : dd. tritt noch der Gradient eines Skalars additiv hinzu. daB die folgenden Differentialgrbfien formationen (2) invariant bleiben. x*') Eigentlich hiitten wir in der ersten Gleichung Konstante mal a. ist deshalb zulassig. x '. K r a m e r s .= dx° + ^ yoo dx\ (4) (5) dS5 = (y. xa'. . Die Vermutung liegt nahe. 1922. Proc. 1 3 £ 3 '. Wenn x° mittransformiert wird. Amsterdam 28. dafi die vier yoi (i r/r 0). das Summenzeichen fortlassen.896 Oskar Klein. Dies bedeutet. daO nur die Verhaltnisse der yik einen physikalischen Sinn kaben. Die Beschr'ankung auf den Wert Eins der Konstante ist ja aber ganz unwesentlich. Hieraus folgt. wo eine an die nun folgenden Betrachtungen erinnernde tiberlogung mit einem einfachen Beweis fur die Invariant von d& und <Zs2 gegeben ist.k — V°ll°*\ dxidXk. sich wie die kovarianten Komponenten eines gewohnlichen Vierervektors transformieren. ^ ^0 0 / In diesen Ausdriicken soil iiber die doppelt vorkommenden Indizes von 1 bis 4 summiert werden. dafi die erlaubten Koordinatentransformationen sicli auf die folgende Gruppe beschranken J ) : x° = x0' + ^(x1'. Die Grofien d %• und ds hangen in der folgenden Weise mit dem Linienelement da zusammen: da2 = ud&2 + ds\ (6) Auf Grund der Iiivarianz von d& und y00 folgt nun. Wie man leicht zeigt. yik nicht von der fiinften Koordinate x° abhangen. 2. (2) x* — rl>i(x '. 3. Bei solchen Summen wollen wir. 7. wenn x° festgehalten wird. Nr. Indem wir die Mafleinheit von x° vorliiufig unbestimmt lassen. A. x*'). (i = 1. x*'. setzen wir: no = »• (3) bei den Trans Man zeigt femer. Dann ist diese Annahme nur eine immer mogliche Konvention. H. Die Annahme »>oa. bleibt y 0 0 bei den Transformationen (2) invariant. dan die GroDen: d x>: d x* ') Vgl. 4).
Wir nehmen deshalb an: (19. yoi = afitpt (* = 1. Die Differentialform ds wollen wir mit dem Linienelement der gewohnlichen Relativit'atstheorie identifizieren. Auf dieses schwierige Problem wollen wir hier nicht naher eingehen. und wo die q>c so definiert sind.= d.^ r • 1 (13) K In dem Ausdruck von P denken wir uns. dafl die gewohnlichen Feldgleichungen von derp Gesichtspunkt der fiinfdimensionalen Geometrie sich einfach zusammenfassen lassen. dafl in rechtwinkligen G a l i l e i s c h e n Koordinaten gilt: {<Px. dafl in rechtwinkligen G a l i l e i s c h e n els2 = dxi + dyi + dzi — cidti. 59* \ dx ^ dx s dx> ) ' . h. Die Groflen yoi verhalten sich also vom invariantentheoretisc'hen Gesichtspunkt wie die elektromagnetischen Potentiate cpi. Wir bilden die Invariante: i (i\ ^ \i k 2 tk wo y ik _tl dx>< _ All dsr + . daB sich fur die gik und (»j in geniigender Annaherung die Feldgleichungen der gewohnlichen Relativit'atstheorie ergeben. sondern wir wollen nur zeigen. (11) Hiermit sind die Groflen yik auf bekannte Groflen zuriickgefiihrt. ) die C h r i s t o f f e l s c h e n Dreiindizes u symbole bezeichnen. und daB y00 = a ist. f» V\ fM _ ft b] \{iv\ (12) . 1 v\ \fi J \p)\v J die kontravarianten Komponenten des fiinfdimensionalen metrischen Fundamentaltensors sind und wo \ . 9>J/I qos) = c v A dx° f Pyidx1. V das gewohnliche skalare Potential und e die Lichtgeschwindigkeit bezeichnen. Wir setzen also wobei wir die gik Koordinaten gilt: Yik = 9ik + « P ft <pk > (10) so wahlen wollen. 4). also: rs = iV 2 ^ f» fe + ^ Y . 2. dafl alle Groflen von x° unabhangig sind. Das Problem ist nun. (7) > q>z = — < wo A das gewohnliche Vektorpotential. solche Feldgleichungen fur die Groflen yik aufzustellen. (8) wo j3 eine Konstante bedeutet.Quantentheorie unci funfdimensionale Relativitiitstheorie. 897 sich wie die kovarianten Komponenten Fik des elektromagnetischen Feldtensors transformieren. 3.
P a u l i . wo y die Determinante der yik bedeutet. f = *. Das Variationsprinzip: 8J=0 fiihrt dann zu den folgenden Gleichungen: JS" und 4 9ikB .4). g die Determinante der \gik und schlieBlich F'l1 die kontravarianten Komponenten des elektromagnetischen Feldtensors bedeuten. (16 b) wo R die E i n s t e i n s c b e Kriimmungsinvariante. „ . Rik die kontravarianten Komponenten des E i n s t e i n s c h e n Kriimmungstensors.3. Sik die kontravarianten Komponenten des elektromagnetischen EnergieImpulstensors. (17) wo % die von E i n s t e i n gebraucbte Gravitationskonstante bedeutet. gik die kontravarianten Komponenten des E i n s t e i ' n s c h e n Fundamentaltensors.898 Oskar Klein. Wenn wir uns auf die in der Elektronentheorie und der Relativitatstheorie iibliche schematische Behandlungsweise derj Materie beschranken. Hierbei soil a als eine Konstante betraclitet werden. kOnnen wir die gewOhnlichen Gleichungen fiir den nicht materiefreien Fall in ahnlicher Weise erhalten. ' '. Relativitatstheorie.. „. ns . wollen wir erst den auf ein Elektron oder einen Wasserstoffkern beziiglichen Tensor: dxl dxk dl dl ] ) Siehe z. _ I ^ L _ = 0 (15) + ° ^ " S** = ° ( * .2. wobei (14) deren Randwerte nicht veiiindert werden sollen. so sehen wir. 719 und 724. Betrachten wir nun das iiber ein geschlossenes Gebiet des ftinfdimensionalen Raumes ausgefiihrte Integral: J = ^P)/~^ydxadx1dx*dx3dxi. Wir ersetzen P in (14) Um die ®ih zu definieren. * = ! . Setzen wir „. S. W i r bilden SJ durch Variieren der Grofien yik und . 3. B. 4) (16 a) ^=1. 2. W. daG die Gleichungen (16 a) in der Tat mit den Gleichungen der Relativitatstheorie fiir das Gravitationsfeld und (16 b) mit den generalisierten M a x w e l l s c h e n Gleichungen der Relativitatstheorie identisch sind fiir einen materiefreien Feldpunkt*).
Aus den Feldgleichungen folgen natiirlich auf gewohnliche Weise die Bewegungsgleichungen fiir materielle Teilchen und die Kontinuitatsgleichung. = £. Die in Frage stehenden Gleichungen folgen nun durch Divergenzbildung von (23). Elektron. Wie man leicht sieht. Wesentlich ist hier nur — = const. Wie man l ) Die speziellen "Werte von — sind natiirlich in diesem Zusammenhang ohne Bedeutung. Weiter ist dx = — V— ds* das Differential der Eigenzeit. die den Bedingungen (19) und (20) geniigen'). M und m die Massen von Wasserstoffkern bzw. aI . konnen von unserem Staudpunkt aus einfach zusammengefaBt werden. Dabei gilt das obere Wertsystem fiir den Kern. 899 betrachten. Wir kommen dann wieder zu Gleichungen vom gewohnlichen Typus. 4.~ 2 Yin ~dT (21) Geschwindigkeits die kovarianten Komponenten des fiinfdimensionalen vektors v' sind. das untere fiir das Elektron. 2. e t (19) '° dl dx Jl ai wo allgemein Vi  \y m \\j m I (20) dx>u ft. die mit den gewbhnlichen Feldgleichungen identisch werden. aber nicht beide Null). 3. Die Rechnungen. die dazu fiihren. und dl ein gewisses invariantes Differential bedeutet. k = 0. wenn wir setzen: dx . Die ®ik sollen gleich der auf die Volumeneinheit bezogenen Summe der^'* fiir die verschiedenen Teilchen sein. dafl sich die elektrischen Teilchen auf fiinfdimensionalen geodiitischen Linien bewegen.Quantentheorie und funfdimensionale Relativitatstheorie. Hieraus folgt. 1. 'i:. wo die Pik die kontravarianten Komponenten des verjiingten fiinfdimensionalen Kriimmungstensors sind (den Hik entsprechend). wo dx' die Lageanderungen des Teilchens bezeichnen. sind namlich unsere Feldgleichungen mit den folgenden 14 Gleichungen Equivalent: pik _ i yikp _j_ % &ik — o (23) (i. wo . Ferner bedeuten e das elektrische Elementalquantum.
§ 2. Immerhin mochte es nicht olme Interesse sein. die gewisse Funktionen der Koordinaten sein sollen. die in der neueren Quantentheorie zum Vorschein kommen. weil x" in den yik nicht vorkommt. in Beziehung zu der funfdimensionaleii Relativitatstheorie zu bringen. Wir kommen dazu. dafi sich siimtliche 14 Feldgleichungen in so einfacher Weise vom Standpunkt der Theorie von K a l u z a zusammenfassen lassen. die Theorie der stationaren Zustande und die damit zusammenhangenden charakteristischen Abweichungen von der Mechanik. Wir bekommen dann: ^ 0 * * ^ .900 Oskar Klein. ^ 7 = 0. Setzen wir » = 5*> (27) so kbnnen die Differentialgleichungen der Strahlen bekanntlich in der folgenden H a m i l t o n s c h e n Form geschrieben werden: d * dH dx' = d dH dpi ^ = dX. sind diese Bedingungen eben deshalb mit den Gleicliungen der geodatischen Linien vertraglich. Betrachten wir erst eine durch (24) bestimmte Wellenausbreitung. wenn wir setzen: u = Acia. daU in (24) nur die mit ca2 proportionalen Glieder berucksichtigt zu werden brauchen. die sich auf unseren funfdimensionalen Raum beziehen soil und als eine einfache Verallgemeinerung der Wellengleichung betrachtet werden kann: Hier bedeuten die aik die kontravarianten Komponenten eines fiinfdimensionalen symmetrischen Tensors.J a c o b i s c h e n partiellen Differentialgleichung der Mechanik entspricht.f (25) und ca als so grofi annehmen. D i e W e l l e n g l e i c h u n g d e r Q u a n t e n t h e o r i e . Die Gleichung (24) besteht unabhangig vom Koordinatensystem. Betrachten wir zu diesem Zweck die folgende Differentialgleichung. Wir gehen nun dazu iiber. die dem Grenzfall der geometrischen Optik entspricht. (28) . dafi wohl keine geniigenden Grunde fur die exakte Giiltigkeit der E i n s t e i n s c h e n Feldgleichungen vorliegen. Es nmfl hier daran erinnert werden. (26) eine Gleichung. die der H a m i l t o n . sofort sieht.
ergibt sich durch den Umstand.Quantentheone und fiiafdimensionale Relativitiitstheorie. . Die Strahlengleichungen lauten nun: & = «• <3S*> dPi _ 1 dgfiV dardx' 7l2TxTWJl+(ip<>^dX Aus f<.3. dx^ C ftd^2 — c 2 d t 2 = 1 . dafi die Strahlen als geodatische Nullinien der Differentialform: ~^>"] aikclxidxk betrachtet werden kbnnen. dxi "I^+PPM dl 0=1. (35) Tk wo ux .. dafi unserc Strahlengleichungen mit den Bewegungsgleiohimgen elektrischer Teilchen identisch werden. = aikdxidxk = ^(dQf f ds\ . ^ ' 4 )  (Sb» 0 . Setzen wir.* = » .4).. ui die kovarianten Komponenten des gewbhnlichen Geschwindigkeitsvektors bedeuten. . on (32) so kbnnen wir durch passende' Wahl der Konstante fi crreichen.2. um dies einzusehen: 1 so folgt (cl&\* 1 /ds\* dL d& (34) Td^ = ^^l und dL . wo die aik die zu den aik reziproken Groflen bedeuten.d#2 + ds 2 = ergibt sich dcpf. 0. die der L a g r a n g e s c h e n Form entspricht. (29) (30) Eine andere Darstellung dieser Gleichungen. 901 H=  2 Aus (26) folgt noch 11= a'^ l f t . also 2 betzen wir nun 2 «.
um zu den gewbhnliclien Bewegungsgleichungen zu +c c 0*0 = e c Aus (37) ergibt sich dann: /3 2 ilfV P fiir den Wasserstoflkern. so wahlen. Insbesondere sind die nach (35) definierten Grbflen pt identisch mit den auf gewbhnliche Weise definierten generalisierten Momenten. annehmen: IM fiir den Wasserstoffkern. Po =  — 1 fiir das Elektron. Dies ist ja wenig bef riedigend. die sich gewissermaBen wie die positiven und negativen elektrischen Teilchen zueinander verhalten. ft*TO2C * Die Gleichungen (35). kbnnen wir A — (l A it r. (36) stimmen dann mit den gewOhnliclien Bewegungsgleichungen elektrischer Teilchen in Gravitationsfeldern und elektromagnetisohen Feldern vollst'andig iiberein. f* 1 fiir das Elektron. daC man bei einem einzigen Werte von (i zwei verschiedene Klassen von Strahlen erhalt. Die Tatsache aber. und 1 fiir den Wasserstoffkern. wollen wir setzen: 0 = ±c (41) Es ergibt sich dann einfach: I \. fiir den Wasserstoffkern. . dafi d% dl gelangen. Da wir (i noch beliebig wahlen kbnnen. (38) Ferner miissen wir. (39 a) (40 a) Wie man sieht. Da nach (34) und (36 a) — und also auch — konstant ist. miissen wir in (37) fiir die Quadratwurzel das positive Zeichen im Falle des Kerns und das negative Zeichen im Falle des Elektrons wahlen. \m fiir das Elektron.1 fiir den Wasserstoffkern. (39) fiir das Elektron. kbnnte als ein Hin .902 Oskar Klein. was fiir die folgenden Uberlegungen wichtig ist. (40) fiir das Elektron.
2]t. Es wird dann eine stehende Schwingung moglich sein. von denen der eine Teil nur von einer einzigen Koordinate. x2. dafi sich die Bewegungsgleichungen beider Arten von Teilchen aus einem einzigen Wertsystem der Koeffizienten ergeben. so ergibt sich also die gewohnliche Quantenbedingung fiir eine separierbare Koordinate. wo sich <£ in zwei Teile spalten lafit. Wegen der Erhaltung der Phase ist die Bedingung dafiir einfach: a)<bpdx wo « eine ganze Zahl bedeutet. die Wellengleichung so abzuandern. Es folgt namlich: d<t> ^ dOM«' ^ dU Die Phase wird also von der Welle mitgefiihrt. sondern wir wollen dazu iibergehen. die mit jder Zeit periodisch bin und her schwingt. sagen wir x. daB [die Wellenausbreitung nach den Gesetzen der geometrischen Optik vor sich geht. Auf diese Frage wollen wir jetzt nicht weiter eingehen. wenn wir die mit der Wellengleichung vertraglichen.Quantentheorie und Jiinfdimensionale Relativitatstheorie. (45) wo /i die P l a n c k s c h e Konstante bedeutet. der sich sofort aus (28 und (30) ergibt. Ahnliches gilt natiirlich fiir ein beliebiges Periodizitatssystem. und dabei annehmen. miissen wir nach (27) setzen: 0 = — x° + S(x\ x2. dafi eine in einem gewissen Augenblick durch (25) dargestellte harmonische Welle nach einer Periode von x mit derjenigen Welle in Phase zusammentrifft. = T . Dazu bediirfen wir des wohlbekannten Satzes von der Erhaltung der Phase. die aus (32) folgende Wellengleichung im Falle des Elektrons etwas naher zu betrachten. Da fur das Elektron pQ = — 1 angenommen wurde. x3. (42) Die Theorie von de B r o g l i e ergibt sich nun. Die gewbhnliche Quantentheorie der Periodizitatssysteme entspricht also vollstandig der Behandlung der Interferenzerscheinungen mittels der Annahme. Betrachten wir nun den einfachen Fall. (44) Setzen wir: 2jt «. die sich aus derselben Lbsung (25) durch Einsetzen der neuen Werte von x°. xa. einem bestimmten Wert von a entsprechenden stehenden Schwingungen aufsuchen. 903 weis darauf angesehen werden.4). x* ergibt. die dadurch charakterisiert wird. a. abhangt. = n. dafi es vielleicht moglich ist. dafi .
= hv = mc2 4. in welcher in (48) die Zeit auftritt. dafl dieselbe in dem Fall. c. Heisenberg. sich die Wellen nach den Gesetzen der geometrischen Optik ausbreiten. (49) [(/»v — e V)2 — m2 c4] i\.~ Setzen wir noch: ( h HI ' 4> (x. 1.904 Oskar Klein. wo a nicht so grofi ist. in Ubereinstimmung mit (42) und (45) ansetzen: x° \ .— Dies in (48) eingefiihrt. Als eine Stiitze fiir diese Form der Quautengleichung kann angefiihrt werden. wo V harmonisch von der Zeit abhangt. deren stebende Schwingungen bekanntlich Werten von E entsprechen. Dann liaben wir in kartesiscben Koordinaten: dfr = da. Wir bescbriinken uns dabei auf den einfacben Fall eines elektrostatischen Feldes.^ p*.E. wie man durch eine einfaclie Storungsrechnung zeigen kann. Fiir U konnen wir..dz2 — c2dt2. die mit 1 ) AuBer durch das Auftreten von x°. ergibt: z/ 1/. Betrachten wir nun auch die Gleichung (24) in dem Falle. } (4b) ds a = dx* 4. 2 ) S c h r b d i n g e r .. 0. z abhangt. (50) (51) so bekommen wir die von S c h r b d i n g e r 2 ) gegebene Gleichung._. (47) In der Gleichung (24) konnen wir nun die mit  ' 1 proportionalen GrbCen vernachlassigen. 4. AV. da V nur von x.0 — eVdt. uuterscheidet sich diese Gleichung von der Schrodingerschen Gleichung durcli die Art. y. dafl wegen (42) die Beziehungen (44) i (45) bei den Koordinatentransiormationen (2) invariant bleiben. das ja fiir die Anwendungen unwesentlich ist. Es mag noch hervorgehoben werden. die sich in ahnlicher Weise zu der Dispersionstheorie von Kramers verhalten wie die Schrodingerschen Losungen zu der Quantentheorie der Spektrallinien. z). daC wir nur die iu a quadratischen Glieder zu beriicksicbtigen braucben. Also ergibt sich H = 1 x (pi +14 1 + P?) lT1 (Pt + e VpX in* r2 + . denn die Dreiindizessymbole sind in diesem Falle nach (17) kleine mit der Gravitationskonstante x proportionale GrbCen. y. Diese Beraerkung verdanke ich Dr. .ay* 4. .. Wir bekommen also 1 ): 1&U 2cV d2U i 4 2 e2V\d2U . Losungen besitzt.
wie wir sahen. In dieser Weise wird man zu der Theorie von K a l u z a gefiihrt. Demi nach dieser Hypothese ist es ja nicht so befremdend. Denn es ist bekanntlicb immer weniger wahrscbeinlich . Denkt man sich aber die beobachtete Bewegung als eine Art Projektion auf den Zeitraum von einer Wellenausbreitung. die genannte Analogie zu dieser Vorstellung in Beziehung zu bringen. Obwohl die Einfiihrung einer fiinften Dimension in unsere pbysikalischen Betrachtungen von vornherein befremdend sein mag. wohl aber einer fiiiifdimensionalen Wellengleichung aufgefaBt werden kann. welche das Gravitationsfeld und das elektromagnetiscbe Feld bestimmen. nach (51).J a c o b i s c b e Gleichung nicht als Charakteristikengleichung einer vierdimensionalen. daB die Bewegung der materiellen Teilchen Ahnlichkeiten aufweist mit der Ausbreitung von Wellen. Die Frequenzbedingung besagt. so lafit sich. solange man eine Wellenausbreitung in einem Raum von nur vier Dimensionen betracbtet.Quantontheorie und fiinfdimensionale Relativitatstheorie. wie S c h r o d i n g c r hervorgeboben hat. Wie man sieht. daB die zti dem System gehbieiulen Lichtfrequenzen den aus deu verscbiedenen Wcrten der Frequenz v gebildoten Differenzeii gleich sind. die Analogie zwischen Mechanik und Optik. Wie die Arbeiten von de l l r o g l i o sind obenstehende Uberlegungen aus dem Bestreben entstanden. Mathematisch ausgedrtickt heifit dies. daB die H a m i l t o n . daB die materiellen Teilchen als spezielle Losungen der Feldgleichungen aufzufassen sind. ist E in dem Grenzfall der geometrischen Optik gleich der auf gewohnliche Weise definierten mechaniscben Energie. scheint ja die Ahnlichkeit der Bedingungen fur die stationaren Zust'ande von Atomsystemen mit den Interferenzerscbeinungen der Optik anzudeuten. die Analogie vollstandig machen. Nun steben bekanntlicb Begriffe wie Punktladung und materieller Punkt schon der klassischen Feldphysik fremd gegeniiber. Auch wurde ja ofters die Hypothese ausgesprochen. Dies kommt schon in der variablen Gescbwindigkeit der materiellen Teilchen zum Vorschein. wird eine radikale Modifikation der den Feldgleichungen zugrunde gelegten Geometrie doch wieder in ganz anderer Weise durcb die Quantentheorie nahegelegt. fur ein tieferes Verstandnis der Quantenerscheinungen auszunutzen. die in der H a m i l t o n s c h e n Methode zum Vorschein kommt. § B. Die in Rede stehende Analogie ist jedoch unvollstandig. 905 den aus der H e i s e n b e r g s c h e n Quantentheorie berechneten Energiewerten identisch sind. die in einem Raum von fiinf Dimensionen stattfindet. S c h l u l i b e m e r k u n g e n . Es liegt nahe. Dafi dieser Analogie ein reeller physikaliscber Sinn zukommt.
. . wie in dem Mesigen Institut fur theoretische Physik beschaftigt. als ganz provisorisrl. Auch wird die Frage ganz offen gelassen. oder ob die S c h r o d i n g e r s c h e Methode die Einfuhrung einer neuen ZustandsgroGe bedeutet. Jedenfalls mufl betont werden. N. daC die Quantenerscheinungen eine einheitliche raumzeitli. haben einen entschiedenen Einflafl auf das Entstehen der vnrliegenden Note gehabt. R a n d a l l und Prof. dafi die zwei Arten von elektrischen Teilclicn durch verschiedene Gleichungen vom S c h r o d i n g e r s c h e n T y p u s behandelt werden.. Dies kommt wohl besonders in der auf S. H. Ob hinter diesen Andeutungen von Mciglichkeiten etwas Wirklicbes besteht. wogegen die Mbglichkeit.Q die in dies. sowohl was die Feldgleiehungcn a)auch die Theorie der stationaren Zustande betrifft.Note versuchte Behandlungsweise. 89* ) T . Ich mbchte auch an dieser Stelle Prof. geworden. Aim Arbor. da. muB natiirlich die Zukunft entscheiden. 1. ob man sich bei der Beschreibung der physikalischen Vorgange mit den 14 Potentialen 1 n_ gniigen kann. Bohr bei mehreren Gelegenheit™ gemacht hat. zu betrachten ist. Quantentheorie und fiinfdimensionale Relativitatstheori<. Mit den in dieser Note mitgeteilten Uberlegungen habe ich micli sowohl in dem Physikalischen Institut der University of Michigan. diese Erscheinun^f. sowie in dem ITmstand. 2 ) Bemerkungen dieser Art. Bohr meinen warmsten Dank aussprechen.. Beschreibung zulassen. 2A. durch ein System von funfdimensionalen Feldgleichungen darzustelkii wohl nicht von vornkerein auszuschlieiten ist 2 ).75 906 Oskar Klein. die Prof. wahnten schematischen Behandlungsweise der Materie zum Vorselicin.
Akad. (1) where here. de Broglie. The equations of motion of electrically charged particles then take the form of geodesic equations. We write this as: dcr = \/y^.2.likdxidxk. Ann.x*') (i=l. which corresponds closely to the theory of Kaluza. 3 E.0. Bd.x2f. Ann. L.Quantum Theory and fivedimensional Theory of Relativity By Oskar Klein in Copenhagen. First of all. one is lead almost automatically to a second order partial differential equation. If one then considers solutions to these equations which depend harmonically on the fifth coordinate.4). Abstract On the following pages I would like to point out a simple connection between the theory suggested by Kaluza 1 on the connection between electromagnetism and gravity and the method suggested by de Broglie 2 and Schrodinger 3 for the treatment of problems in quantum theory. \ x = 4>i(x1'. the quantities 7. Paris 1924. Kaluza. the symbol J ] denotes s u m m a t i o n from 0 to 4 over repeated indices.x4. T h e 15 quantities 7. 1 Fivedimensional Theory of Relativity I begin with a brief account on the fivedimensional theory of relativity. F r o m this follows that the allowed coordinate transformations are restricted to the following group: x0 = x0' + M^'^2'^3'^4'). Zeitschrift fur Physik. Theses. 79. (10) 3. Let us consider a fivedimensional R i e m a n n i a n line element. Sitzungsber.fc need t o be independent of the fifth coordinate a. 1925. Schodinger.x3. for which we postulate a coordinate independent meaning. Phys. four of the coordinates. p. have to characterize the usual spacetime. To get from these to the quantities gtk and fa of the ordinary theory of relativity. Phys. we need to m a k e some special assumptions. 1926. d. which has as well as the usual four dimensions an additional fifth dimension. Berl. with a period depending on Planck's constant.fc are the covariant components of a fivedimensional s y m m e t r i c tensor.April 1926. and in the following. 2 . 895906. XXXVII.x3'.22.k of a line element of a Riemannian space. which can be regarded as a generalization of the usual wave equation. say x1. The theory of Kaluza essentially combines the ten Einstein gravitational potentials gtk and the four electromagnetic potentials rj>i a s the coefficients f.3. p. d. one obtains the above mentioned quantum theoretic methods. 361 and 489. 1921. Here x° • • • x4 denote the five coordinates of the space. J i (o) K> 'Th. but differs from this in several points.x2. d. Received on 28. 966. Secondly. If one interprets these as radiation equations by considering matter as a kind of wave propagation.
Then the above assumption is simply a choice of convention. with a simple proof of the invariance of dd and ds2 .<S>z) = A . V the usual scalar potential and c the speed of light. as usual. When x° is transformed as well. behave from the point of view of invariance theory like the electromagnetic potentials </>. a gradient of a scalar needs to be added. Proc. leave out the summation sign.3. 1922. Therefore the assumption 700 =const. H.4). J where A is the usual vector potential. (5) V Too / In these expression the summation is carried out over the indices 1 to 4. the quantities 70. we shall. \ . (4) 7oo =(lik^^\dxidxk. We would like to identify the differential form ds with the line element of ordinary relativity. Nr. A m s t e r d a m 2 3 .^.= a/% (» = 1. where a thought similar to the considerations below is presented. The problem is now to find field equations for 7. Hence. It is natural to presume that only the ratios of 7 ^ have a physical meaning.<i>y.In fact in the first equation we should have written constant times x But the restriction to the value 1 of this constant is immaterial. For now we leave the unit of x° undetermined and we set: 7oo = a(3) Further one can show that the following differential quantities remain invariant under the transformation (2). is allowed. l9) fi<f>idxi. when x° is fixed. that 700 is invariant under the transformation (2). This means. which it is always possible to make. and the fa are defined in such a way that in orthogonal Galilean coordinates the following holds: {<l>x. The quantities dd and ds are related to the line element da in the following way: ds 2 da2 = ad62 + ds2. 7o. We therefore assume: d9 = dx° + i. Therefore we set: l i k = 9ik + af32</>i<f>k.e. (7) 4>o = cV. instead of x' One can easily prove. namely 4 : d0 = dx°+^dxi. transform like the components of an ordinary fourvector. (8) where /3 is a constant. For such sums. have been expressed in terms of known quantities. A.. (6) It follows from the invariance of d6 and 700 that the four f0i (i ^ 0).^ such that these reduce to the equations for go.2. (11) Thus the 7. 7. Kramers. and <f>i of 4 Cf. that the quantities: djoi _ d/ok dxk dx' transform like the covariant components Fas of the electromagnetic field tensor. (10) where we have chosen gik in such a way that in orthogonal Galilean coordinates: ds2 = dx2 + dy2+ dz2c2dt2.
If we set: = K. as used by Einstein.g.ordinary relativity in a sufficiently accurate approximation. we shall first consider the tensor for one electron or one hydrogen nucleus: eik = dl dl 5 d ^d4 (is) S e e e. We only want to show.k = 1. p .} (12) where 71* are the contravariant components of the fivedimensional metric tensor and { r ' } are the Christoffel symbols. g'k the contravariant components of the metric tensor. S the contravariant components of the electromagnetic energymomentum tensor. (14) where 7 denotes the determinant of 7. We then compute 8 J by varying 7. R'k the contravariant components of the Einstein curvature tensor.2. and yoo = a. If we restrict ourselves to the schematic treatment of matter. We shall not investigate this difficult problem further. we see that indeed equation (16a) is identical to the equations of relativity for the gravitational field and (16b) is identical to the covariant form of Maxwell's equations in the absence of matter.4) (16a) dV^F =0 (1=1. (166) where R is Einstein's curvature invariant.Pauli.3.0.4). which is evaluated over a closed region of the fivedimensional space: J= J P^^dx°dx1dx2dx3dx4.3. 719 and 724 .2.^ and g^fS leaving the boundary values fixed. that from the point of view of the fivedimensional theory. Let us now consider the following integral. W. The variation principle: SJ = 0 imp ies the following equations: Rik\gikR and „ + ~2S'k=0 p u a/3 {i. We replace P in (14) by P+ Kj2fikQik To define Q'k. that is: In the expression for P we regard all the quantities as being independent of a.. We form the invariants: £f fc dxk dxf \ v ) \ fi J \ fj c. g the determinant of gik and finally F'^ are the contravariant components of the electromagnetic field tensor. we can obtain in a similar way the usual equations when matter is present. Relativitatstheorie. as it is common in the relativistic theory of electrons./. the ordinary field equations can be combined in a simple manner. . 10 2 where K is the gravitational constant.
3.4.Here dx' is the change in position of the particle and dl denotes a certain invariant differential. W Also. It can easily be seen that our field equations are equivalent to the following 14 equations: pik _ iykp + KQik = Q ( 23 ) (i. which obey the conditions (19) and (20) 6 . The equations in question can now be obtained by taking the divergence in (23). because the 7. It is immediate. if we set: ^i= ±h 5 =1^' where generally (19) (2°) (21) dl [ \Jm *= I>^r is the covariant component of the fivevelocity vector dr' " = if ' dr. Further V^di? is the differential of the proper time. that electrically charged particles move along fivedimensional geodesic lines. and thus connect the characteristic deviations from mechanics. M and m are the masses of the hydrogen nucleus and electron. The Q'k have to be equal to the sum of the 9'k for the various particles. For this purpose. where P'k are the contravariant components of the contracted fivedimensional curvature tensor (corresponding to the Rlk). in a unit volume. Of importance is here only . which is formulated for our fivedimensional space and which can be considered as a simple generalization of the wave equation: \^ ik ( 6 d2U v^ fi k\ dU\ Obviously. are irrelevant in this context. The field equations imply in the usual way the matter equations of motion and the continuity equations. respectively. which lead to these. e is the fundamental electric quantum. but not both zero). which reduce to the ordinary field equations. which appear in modern quantum theory. the upper value denotes the term for the nucleus and the lower the one for the electron. are independent of x°. from the point of view of Kaluza's theory. that these conditions are compatible with the geodesic equations. with fivedimensional relativity. It is certainly of some interest that. the particular values of ^ £ =const. let us consider the following differential equation. 2 The Wave equation of Q u a n t u m Theory We will now give the theory of stationary states. k — 0. It is necessary to remind the reader that there is not sufficient evidence to show that Einstein's field equations hold exactly. we obtain equations of the ordinary type. these 14 field equations can be summarized in such simple fashion. Again. Here.1. From our point of view we can now simply summarize the calculations. 2.7. It follows.
§. To see this.„. The ray equations then become: ^ = 0.= «. Lagrangian. (35) (36a) . If we set: « = 5?' (27) we can write the differential equations for the rays in the following Hamiltonian form: i dx* = £^A. (26) implies H = 0. dX 9%t where u\ • • • M4 are the usual covariant components of the velocity vector. (30) A different. (34) and Pi = 5 .2.4).80 Here. We obtain this by setting: « = Aeiu>* 2 (25) and taking u so large that we have to consider only terms proportional to LJ in (24). a'k are the contravariant components of a fivedimensional symmetric tensor. which are functions of the coordinates. we can make our ray equation coincide with the equations of motion of electrically charged particles. Equation (24) holds independently of the choice of coordinate system. Let us begin by considering the wave evolution given by (24).3. (3D (32) and suitably choosing the constant ji. form of this equation can be obtained using the fact that the rays can be viewed as geodesic null lines of the differential form: y aikdx'dx where a^ are the dual quantities to a'k.. > We then obtain: This equation corresponds to the HamiltonJacobi partial differential equation of mechanics.S + '8Po^ (1 = 1. we set: L which implies r 1 (de\2 1 fds\2 = 2" U J + 2 U J > Po (33) di de = jai = ^dJ dX . dpt (28) where Further. that is £"*«*" = 4 = {i. in the special limit of geometric optics.?/ Setting ] T aikdxidxk = n(de)2 + ds2.
. J  and so also ^ are constant. thus assuming that the wave evolution proceeds according to geometric optics. Nevertheless.. . so we can choose A such that dr dX ( M for the hydrogen nucleus.„ . the quantities p. we further have to assume that: . f . which in some sense relate to each other like positively and negatively charged particles. . We can still choose /?. In particular. it follows that: . (42) The theory of de Broglie is obtained by finding the standing waves. (36) then exactly coincide with the usual equations of motion of electrically charged particles in gravitational and electromagnetic fields. 1 m for the electron. .x4). { and (39a) . Indeed. 0P° = \ C _ i for the electron.c2dr'2 = 0 it follows that fif = crfi.i . so we shall set: ?=.1 for the electron. To obtain the usual equations of motion. M c i J & ' (40a) —srm for the electron.x3. Since we assumed p0 = — 1 for the electron.d£ d\ From Id^Wd* 2 dx' d\ d\ *^d£ ox1 d\ (! = 1 . 7 " ~"° ' a . . (27) implies that we have to set: <J> = x° + S{xl. (39) (40) The equations (35). M for the hydrogen nucleus. We need to make use of the wellknown theorem of conservation of phase.x2. ' For the squareroot in (37) we need to choose the positive sign in the case of the nucleus and the negative sign for the electron. It follows from (37) that: it = { ^M2c4 for the hydrogen nucleus. (41) c It then follows that: +1 for the hydrogen nucleus.„„ . as obtained from (32). (37) From (34) and (36a). We turn to a more detailed analysis of the wave equation for the electron. 3 .for the hydrogen nucleus. We will not investigate this question further. 2 . 4 ) (36) tide2 + ds2 = fidd2 . This is not very satisfactory. one could interpret the fact that two classes of solutions for rays appear for one value of //. defined in (35) are identical with the generalized momenta defined in the usual way. This is important for the following reasons. 4 for the electron. as a hint that it might be possible to alter the wave equation in such a way that the equations of motion of both types of particles result from one assignment of values to the coefficients. which can be obtained directly from (28) and (30). e . which are consistent with the wave equation and correspond to a given value of u. .
It may be worth noting. This will allow for a standing wave. A similar analysis obviously holds for arbitrary periodic systems. which for applications is irrelevant anyway. Now setting: (44) ?. phase. by (17). but also in the way in which time enters equation (48). in accordance with (42) and (45): U = e2™(£"t)rP(x.The phase is therefore transported with the wave. (45) remain invariant under the coordinate transformation (2) due to (42). represented at a particular moment by (25) comes into phase with the wave which is obtained from the same solution (25) but with the new values for x°. (46) ds2 . Hence it follows that: H = \{vl +P2y+Pl) ~ ~(Pt + eVp0f + 7 ^pl (47) We can now neglect the quantities proportional to j 'r > in equation (24). x2. because in this case. (50) (51) T h i s equation differs from the Schrodinger equation. 7 = 0. (49) Inserting this into (48) results in: Vi> + ^[{hueV)2m2c4}4> Moreover setting: hi/ = mc2 + E. where n is an integer. the usual quantum theory of periodic systems corresponds just to the analysis of interference.c2dt2.2x. it has solutions which behave in a similar way to Kramer's dispersion theory as Schrodinger's solutions behave with respect to the q u a n t u m theory of spectral lines. assuming that the waves evolve according to the laws of geometric optics. z. . e2V2\ d2U n dx°2 Since V depends only on x. . We shall restrict ourselves to the simple case of an electrostatic field. We therefore obtain 7 „TT 1 d2U 2eV 82U ( .y.dx° . say x. In cartesian coordinates we then obtain: d9 . I owe this remark to Dr. the condition for this is simply: p dx = n. To support this sort of q u a n t u m equation we note. not only in the appearance of x ° . oscillating periodically in time. («) where h is the Planck constant. we obtain the usual quantum condition for one separable coordinate. Let us now consider (24) in the case where u is not so large as to allow consideration of the terms in u squared only. W . these triple indexed quantities are small quantities. we can make the following ansatz for U. after a period of x. proportional to the gravitational constant K. This can be easily seen by perturbation theory. We can say therefore that.z). that the relations (44).y. that in the case when V depends harmonically on time. We shall now consider the case when $ splits up into two parts. Heisenberg. the harmonic wave. One of these parts will depend only on one coordinate.dx2 + dy2 + dz2 . characterized by the property that.eVdt.
and the fact that two types of electrically charged particles are described by two different equations of Schrodingertype. Bohr. as well as at the Institute of Theoretical Physics here. However. as we have seen. by (51).we obtain Schrodinger's equation. Bohr has made on several occasions. Although it might initially appear alien to include a fifth dimension. In mathematical terms this means. then the analogy is complete. one sees that E corresponds to the mechanical energy defined in the usual way. But. appearing in the Hamiltonian method. that the condition on the frequency means. which govern gravitational and electromagnetic fields. of which the standing wave solutions are known to give values for E identical to the values of the energy calculated from Heisenberg's quantum theory. because from this hypothesis it is not at all strange that the motion of matter particles has similarities to the evolution of waves. It seems. Whether there is something real behind these possibilities one will have to investigate in future. in order to obtain a deeper understanding of quantum phenomena. Remarks of this sort. if one thinks of the observed motion of the particle as a kind of projection of a wave evolution in a space of five dimensions onto spacetime. Also. Randall and Prof. give a real physical meaning to this analogy. Already in the classical physics of fields the notions of point charge and of point matter are somewhat alien. As is know. Schodinger has emphasized. We also leave the question unanswered whether in describing physical processes it is sufficient to consider the 14 potentials. N. M. This reveals itself already in the variability of the particle velocity. that the analogy between the conditions on stationary states in atomic systems and the ones in optical interference phenomena. 3 Final Remarks The above considerations arose. occasionally the hypothesis has been put forward that matter particles should be considered as special solutions to the field equations. that the HamiltonJacobi equation cannot be considered as the characteristic equation of a four. I. just like the works of de Broglie.e. had a crucial influence on the development of these notes. this analogy remains incomplete if one restricts oneself to wave evolution in a space of only four dimensions. In the limit of geometric optics. This is apparent in particular from the schematic treatment of matter in the paragraph containing (18). that the frequencies of light corresponding to the system are equal to the difference of the values for v. I worked on the ideas presented in this note while at the Institute of Physics of the University of Michigan. but only a fivedimensional wave equation. from the endeavour to use the relation between mechanics and optics. it seems more and more unlikely that quantum phenomena allow a unified spacetime description. . Relating this idea with the above analogy seems to suggest itself. or whether Schrodinger's method means an introduction of further parameters. In this way one is lead to the theory of Kaluza. Ann Arbor. which Prof. H. I would like to express my warmest gratitude towards Prof. Schrodinger. But the possibility of describing these phenomena in terms of a system of fivedimensional field equations does not seem to be excluded from the start 9 . We would like to emphasize that the treatment presented in this note of the field equations as well as the theory of stationary states has to be regarded as entirely provisional. this radical change in the geometry underlying the field equations is suggested naturally from an entirely different point of view by quantum theory.
1.) lautet: + m% c2 f i (?l2 — <p2) = 0. seine Arbeit im Manuskript zu lesen. 37.J acobiscke Gleichung (H. . Free d ericksz entstanden.) Die Schrbdingersche Wellengleichuug wird als invariante Laplacesche Gleichung und die Bewegungsgleichungen als diejenigen einer geodatischen Linie im fiinfdimensionalen Eaum geschrieben. S p e z i e l l e R e l a t i v i t a t s t h e o r i e .Uber die invariante Form der Wellen. Die L a g r a n g e s c h e Funktion fiir die Bewegung eines geladenen Massenpunktes ist. Die Einfiihrung eines fiiriften Koordinatenparameters scheint uns zur Aufstellung der S c h r o d i n g e r s c h e n Wellengleiohung sowie der mechanischen Gleichungen in invarianter Form gut geeignet. 1926) in Leningrad eingetroffen. 895. um die Gravitation und das elektromagnetische Feld von einem einheitlichen Standpunkte aus zu betrachten. Wegen der Wichtigkeit der Resultate diirfte aber vielleicht ihre Ableitung auf einem anderen Wege (Verailgemeiuerung eines in meiner friiheren Arbeit gebrauchten Ansatzes) von Interesse sein. Juli 1926. in leicht verstandlicher Bezeichnungsweise. Von V. Wiihrend diese Notiz im Druck war. die prinzipiell mit denen dieser Notiz identisoh sind. U — e <p. V. In seiner noch nieht veroffentlichten Arbeit bedient sioh II. L = — m c2 1/ 1 — \ + — 3t. 2 ) Der Verfasser hat mir liebenswiirdigenveise die Moglichkeit gegeben. Anmerkung bei der Korrektur. dem ich auch manche wertvollen Ratschlage verdanke. M a n d e P ) des Begriffes des fiinfdimensionalen Raumes. Phys.und der Bewegungsgleichungen fur einen geladenen Massenpunkt 1 ). (Eingegangen am 30. ist die schiine Arbeit von Oskar Klein (ZS. (2) J ) Die Idee dieser Arbeit ist in einem Gesprache mit Prof. f. (1) und die entsprechende H a m i l t o n . P. in welcher der Verfasser zu Eesultaten gelangt. Took in Leningrad. Der uberziililige fiinfte Koordinateuparameter steht in enger Beziehung zu der linearen Differeutialform der elektroinagnetischen Potentiale.
242. (4) Ferner bleibt die Wir bemerken. F o c k . daJ3 die Koeffizienten der nuUten.V ] (§*)'.V. 1926. Die letztere Transformation laflt aucb die lineare Differentialform e 1 e 5 (%dx + %dy + ILde) d'Sl mdi H dp (6) in c invariant 2 ). 38. Man findet leicht els2 = dx* + dif + da* — c2 dt* + (d'Sl)2 a) *) V. Wir wollen nun die Form Q als das Quadrat des Gradienten der Funktion 1^ im fiinfdimensionalen Raum (_R5) auffassen. Phys. ZS. f. Op Form Q invariant. 227 Analog dem in unserer friiberen Arbeit 1 ) gebrauchten Ansatze setzen wir bier dtp grad W = grad tp dW dj^ ' dp dt n djT dp erhalten wir erne (3) wo p einen neuen Parameter von der Dimension des Wirkungsquantums bezeicbnet. Dber die invariante Form der Wellengleichungen usw. ) < P 9>i c dt f c (5) P — i>! setzt. 2 ) Das Zeicheu d' soil andeuten. wo f eine willkiirliche Funktion der Koordinaten und der Zeit bezeicbnet. . ersten und zweiten Potenz von =— vierdimensionale Invarianten sind. dall d'& kein vollstandigesDifferential ist. Zur S e h r o d ingerschen Wellenmechanik. Fock. Nacb Multiplikation mit (5—) qua dratische Form « = <*•>•F&) V &(«•«. * + [„V + j > . wenn man 1 = 2^ + g r a d / ' .• + . und sucben das entsprecbende Linienelement.
daB diese Koeffizienten von t nicht abhangen. Fock. Die Bedeutung des iiberzahligen Koordinatenparameters p scheint namlich gerade darin zu liegen. Die L a p l a e e s c h e Gleichung in Rt lautet: Sie bleibt. wiirden sich namlich nur um einen Faktor e ch vom absoluten Betrage 1 unterscheiden und folglich (bei sehr allgemeinen Voraussetzungen iiber die Funktion f) die gleichen Stetigkeitseigenschaften haben. und setzt man nxi — {E+mcm *. daB die Koeffizienten der Gleichung fur rpQ im allgemeinen komplex sind. 13. 12. bei Lorentztransformationen und bei Transformationen (5) invariant. fiir welche eine Funktion ip1 existiert. = ^0e ><• (9) setzen 1 ). Aus den soeben angefiihrten Betrachtungen folgt. Phys. f. = « * *i. kimnen wir die Abhangigkeit der Funktion T> von p in Form / eines Exponentialfaktors ansetzen. ebenso wie (7) und (4). und zwar miissen wir. um tlbereinstimmung mit der Erfahrung zu erhalten p •4. die gewissen Endlichkeits. dafi er die Invarianz der Gleichungen in bezug auf die Addition eines beliebigen Gradienten zum Viererpotential bewirkt. Die beiden mit den Vektorpotentialen 31 und 31 = 21 — grad f erhaltenen 2nie Funktionen i/>a und i/). sind dann die B o h r schen Energieniveaus. Die Gleichung fur IJJ0 ist gegeniiber Lorentztransformationen. 1923) bemerkten Bezielrongen im Zusammenhange stehen.und Stetigkeitsforderungen geniigt. daB die Addition eines Gradienten zum Viererpotential keinen Einflufi auf die Energieniveaus ausiiben kann. Da die Koeffizienten der Gleichung (8) den Parameter p nicht enthalten. l ) Das Auftreten des mit der Linearform verbundenen Parameters p in einer Exponentialfunktion konnte vielleicht mit einigen von E.228 V. Nimmt man ferner an. invariant. Es sei hier bemerkt. Diejenigen Werte von E. Schroidinger (ZS. . welche mit der in unserer friiheren Arbeit aufgestellten Verallgemeinerung der S c h r o d i n g e r schen Wellengleichung identisch ist. nicht aber gegeniiber Transformationen (5). (10) so erhalt man fiir i^x eine zeitfreie Gleichung.
nicht aber von der Beschaffenbeit des Massenpunktes ab.reprasentiert.^ d«/ 4.k=l yikd%idxk (11) = 2 9ikdXidxk\^3 2 \i = 1 4i i) • / dx i. 2. 2. 4) die durch ca dividierten Komponenten des Viererpotentials.Stj d« — q>cdt). 3.k — 1 Hier sind die Grbflen gik die Komponenten des E i n s t e i n s c h e n Fundamentaltensors. Die Groflen gik und qt hangen nur vom Felde. 6) (15) Y ylk z alff • . (12) die Grbfle g6 eine Konstante und #6 der iiberzahlige Koordinatenparameter. W e l l e n g l e i c h u n g .und der Bewegnngsgleichungen usw.Ik 9' — g>*ai = 0 (14) (». (16) y6 = 6 w 2 . 4). A. 2. k. die Grbflen <^ (i = 1. bei der Summation von 1 bis 4 dagegen unterdrtickt. 4. 2. Alle Koeffizienten sind reell und von xb unabhangig. Mit diesen Bezeichnungen finden wir Yik ' 9ik + aiO>k\ 9a = y*t = .Ak . 5) (13) 6 einfiihren und folgende Verabredung treffen: bei der Summation von 1 bis 5 wird das Summenzeichen angeschrieben. 3. die letztere wird durcb den Faktor —5. (i. I = 1. 3.(1 + flifl° Die der Gieichung (8) entsprechende "Wellengleichung lautet: . m* abhangigen Grbfien 1i 6^ Zur Abktirzung wollen wir aber die von — m a{ (i = 1. 3.87 Ober die invariante Form der Wellen. also 4 2 <k d Xi = 1 y ( ^ (i x f. k = 1. Fur das Linienelement im funldimensionalen Raum setzen wir an: 5 229 ds 2 = S i.2. A l l g e m e i n e R e l a t i v i t a t s t h e o r i e . 4.
i t i i = 0. Fock. tkl] (20) und spalten den Tensor 2 Aik in seinen symmetrischen u n d antisymmetrischen T e i l : 7i r A . 5 k5 5 55 (5 5 I 5 = 2 a. y g OXiX oxj V. ab oxtox + —. J* L* 2 a. B e w e g u n g s g l e i c h u n g e n . so laJt sich diese Gleichung schreiben: ===• •^{]'—gglk =r^) — V— l c e g * V 1 0 4jt3c2 ( » 2 + e 2 &20i/'o = °(19) B. W i r bezeichnen die fiinfdimensionalen Klammersymbole mit \ \ und die vierdimensionalen mit  Wir fiihren ferner die kovariante Ableitung des Viererpotentials ein: Ak da. (18) Fiihrt man endlich die Funktion tjj0 und die Potentiale qt ein.230 oder.( a * 0" Mil + «< 9ir M ik). . (22) — — a'ikf. 0. ausfiihrlicher geschrieben. Zu diesem Zwecke miissen w i r zunachst die C h r i s t o f f e l s c h e n Klammersymbole berechnen.i(l+fl1«')j^ d*i> dxl K) = 0. Mlk —AltAkl Wir haben dann ikl = kl _dat dak —j^kJJt (21) [*] + o. Wir wollen nun die Bewegungsgleichungen eines geladenen Massenpunktes als diejenigen einer geodatischen Linie in B6 aufstellen. (ci^Mu+a^M^).
j . d'£1 . „+ d 9L\ e d< at dx .t d777 " 7 6 ^ 7 ^ ' 7 7 atdXi 4. die man in der Form y dsVds/ ' schreiben kann. wir unterdriicken noch das Zeichen 4 am Klammersymbol: d a «. so erhalt man eine Gleichung. „ — c 2 dT 2 + (d'.und der Bewegungsgleichungen usw. 1 a"£& f~ e /.+ . Es gilt also d'Q — const (27) ds Multipliziert man (23) mit gra —— und summiert iiber r und a. (29) (30) > d /drV .g"Mtl . Die Gleichungen der geodatischen Linie in ii B lauten dann: 231 d£ ds2 d2x&. a da. Ik I) dxk dx.abdxh. dafl die Gleichung (24) eine Folge von (23) ist.. r*ft pa* (r } i ds as *&. In der speziellen Relativitatstheorie laflt sich die erste dieser Gleichungen schreiben 1 d'ilre ( d%\ dq>l m H H + (3d) 17 T~ d dx . (32) dx* {r } dx dx (it dx Das letzte Glied auf der linken Seite stellt die L o r e n t z s c h e Kraft dar. Fuhren wir in (23) die Eigenzeit als unabhiingige Veranderliche ein. B. .girMii dp = as 1 d'£1 as 0> a (23) (24) •d7 + «I^ 1 „ da* d a .&) 2 7a7b\' "* ' dT) Jz\ = ° .r . 7 7 + V • . = s ds a = \ds. wenn man die Eigenzeit x durch die Formel gikdxtdxk eiufUllrt = —c 2 di. die Linearform a" £1 = (25) Multipliziert man die vier Gleichungen (23) mit a„ die fiinfte (24) mit ab und addiert. Die Gleichung (28) oder (30) ist iibrigens wegen (31) eine Folge von (26). Aus dem Gesagten folgt. wie friiher.89 tiber die invariante Form der Wellen. so kommt der fiinfte Parameter ganz zum Fortfall.. so folgt as wegen der Antisymmetrie von Mik d ( dsXdsds/ dxr dxa\ oder.j i = 0. Hier bezeichnet d'fl. r dx. wir konnen sie also beiseite lassen. 2 0.
Fock.m (37) dxb und fiihren wir statt at die Potentiale qt ein. tiber die invariante Form der Bewegungsgleichungen.— = d W % /l_ dip\* \a6 dxj _ ^—. . T.. setzenwirdas Quadrat des funidimensionalen Gradienten einer Funktion I/J gleich Null. mul3 der Faktor der eckigen Klammer den W e r t 1 haben. gelten kann. . (35) Die Bahnen des Massenpunktes sind also geodatisohe Nullinien im fiinfdimensionalen Raum. . die uns als iVusgangspunkt diente.232 V. Es gilt also at und ds2 = 0. Um die H a m i l t o n . L e n i n g r a d . Oil> ~*** mca. Um Ubereinstimmung mit der Erfalirung zu erhalten. Juli 1926.J a c o b i s c h e Gleichung zu erhalten. (38) gik ^ Z L *L1L _ 2 e c a .r — + c (w + e clt (f) = d Xi dxk a xt die als Verallgemeinerung unserer Gleichung (2).k djp dj^ _ _2_ dj)_ dxt dxk ab dx6 Setzen wir hier . so erhalten wir eine Gleichung dWdW dW 1 2 2 a 0. 24. Physikalisches Institut der Unive'rsitat.dj>_ dxt .
. M a n d e l 1 uses t h e idea of a fivedimensional space to consider g r a v i t a t i o n a n d the e l e c t r o m a g n e t i c field from a unified point of you. and the equations of motion as those of a geodesic. 1926) has arrived in Leningrad.) Abstract The Schrodinger wave equation is written as an invariant Laplacian equation. Freedericksz. Phys. The author was generous enough to give me the chance to read his work in manuscript form. in which the author arrives at results which coincide in principle with the ones contained in this note. in fivedimensional space. their derivation in a different way (a generalization of an ansatz used in my previous work) may be of interest. The redundant fifth coordinate parameter is closely related to the linear differential form of the electromagnetic potentials. 895. the i n t r o d u c t i o n of a fifth c o o r d i n a t e p a r a m e t e r seems well suited for the invariant formulation of b o t h Schrodinger's wave e q u a t i o n a n d the m e c h a n i c a l e q u a t i o n s . To u s . In his as yet u n p u b l i s h e d work. from whom I have also received some valuable advice. *The idea for this work was formed in a talk with Prof. While this note was being printed. Remark during correction. H.On the Invariant Form of the Wave Equations and the Equations of Motion for a Charged Point Mass* By V. f. Due to the importance of these results. however. (Received on 30 July 1926. the beautiful work by Oskar Klein (ZS. V. Fock in Leningrad. 37.
ZS.— (2) + m V + —(A2 In analogy to the ansatz used in our previous work 2 we shall set gvadW gradV>. The latter transformation also leaves the linear differential form 6 6 1 d'Vt = 2 (Axdx mc" + Aydy + Azdz) 4>dt f —dp mc mc (6) V. (5) P= Pi~f where / denotes an arbitrary function of the coordinates and time. + <f>dWs . (4) + m V + —{A2 . dip dp i dW _ j f dt i' dp (3) where p designates a new parameter with the dimensions of the quantum of action.dP.^ J we get a quadratic form . Fock. is ir e . Further.) is L mc (1) (gradW) 2  1  (dWs dt 2e ( JTI/ AgradW '•) = o. We note that the coefficients of the zeroth. . Zur Schrodingerschen Wellenmechanik. in obvious notation. 5.P.— ^ ( A g r a d ^ + ^^l c <9p <9*. f. first and second power of  ^ are fourdimensional invariants. Phys. 242. After multiplication by (  ^ j (5 = ( g r a d V ' ) 2 .+ A. the form Q remains invariant if one puts A = Ai+ g r a d / .v z c c and the corresponding HamiltonJacobi equation (H.^ ( . 1926. 38.1 Special Relativity The Lagrangian for the movement of a charged point mass.
(io) one obtains a timefree equation for 0 i . The fact that the parameter p. we can cast the dependence of the function t / ) o n p in the form of an exponential factor.c2dt2 + {d'rt)2. which is identical to the generalization of Schrodinger's wave equation that we established in our previous work. The Laplacian equation in R$ is 1 ^ c2 dt2 ed^ f c dp \ ldf\ c dt I 2e( c \ 2 2 . might be connected to some relations remarked on by E. which is associated with the linear form. f. those values of E for which there is a function tpi that satisfies The symbol d' shall denote the fact that d'Q. it remains invariant under Lorentz transformations and transformations (5). One easily finds ds2 = dx2 + dy2 + dz2 . the coefficients in the equation for ^0 are complex.tf) dp2 = 0 d2tp (8) Like (7) and (4). We now want to interpret the form Q as the square of the gradient of the function ip in fivedimensional space (i? 5 ). To get agreement with experiment we have to set 4 V = ^oe2m» > (9) The equation for ?/>o is invariant under Lorentz transformations. Schrodinger (ZS.invariant . Consequently. and look for the corresponding line element. 4 3 . Since the coefficients of equation (8) do not contain the parameter p. If one further assumes that those coefficients do not depend on t and sets 0O = e . 1923). 12. m2c2 + (7) M dp 6&1> c dtdpj {A c e2 2 f A2 . Phys. in general. is not a total differential. The significance of the redundant coordinate parameter p seems to lie exactly in this fact: it causes the invariance of the equations with respect to the addition of an arbitrary gradient to the fourpotential. Note here that.¥ ( * w ) V i . but not under transformations (5). appears in an exponential function. 13.
94 certain requirements of finiteness and continuity. The functions 0 i and 0 i . It follows from these observations that the addition of a gradient to the fourpotential cannot have an influence on the energy levels.] . 24 July 1926. [The remainder of the paper consists of a generalization to curved spacetime. obtained using the vector potentials A and A — A — g r a d / respectively. Physical Institute of the University. are the Bohr energy levels. 2 General Relativity Leningrad. would only differ by a factor of e~^~s from the absolute value 1 and would therefore (under very general assumptions about the function / ) have the same continuity properties.
t) well inside its tube). revised manuscript received June 16.1959.finally.t) is a solution of the Hamiltonian Ho. the potentials cannot be eliminated from the basic equations.timedetermining electrical "shutters" the beam is tials in the quantum theory. University of Bristol. VOL. for the region inside the cage. the vector and scalar potentials were first introduced as a convenient mathematical aid for calculating the fields. before the electron comes near the 485 . It is true that in order to obtain a classical canonical formalism. then the solution for H will be *=*oeiS/*. Bristol. but differently in each tube. Nevertheless.1959) In this paper. H—Ho+V(t) where Ho is the Hamiltonian when the generator is not functioning. We shall then discuss possible experiments to test these conclusions. time only. England (Received May 28. The potential then grows as a which is.95 THE PHYSICAL REVIEW O/I journal of experimental and theoretical physics established by E. the potentials themselves have no independent significance. Finally.combined to interfere coherently at F. the fundamental equations of motion can always be expressed directly in terms of the fields alone. to no change in any physical result. In this paper. This will the potential is zero in region I (until each packet is add to the Hamiltonian of the particle a term V{x. and as a result. are all gauge invariant. Wills Physics Laboratory. POSSIBLE EXPERIMENTS DEMONSTRATING each part is then allowed to enter a long cylindrical THE ROLE OF POTENTIALS IN THE metal tube. 3 AUGUST 1. 1. there exist effects of potentials on charged particles. with the wavelength X.we shall suggest further possible developments in the interpretation of the potentials. 1959 Significance of Electromagnetic Potentials in the Quantum Theory Y. even in the region where all thefields(and therefore the forces on the particles) vanish. I assume this almost everywhere in the following discussions) we have. We shall begin with a chopped into wave packets that are long compared simple example. and V(t) = e<t>(t). so that it may seem that even in quantum mechanics. In the nonrelativistic limit (and we shall it falls back to zero. contrary to the conclusions of classical mechanics. Nichols in 1893 SECOND SERIES. If ^a{x. INTRODUCTION N classical electrodynamics.)eiS'» = [ffo+F(0>=^. as well as the physical quantities. these equations. the canonical formalism is necessary. L. and. 1. 115. H. By means of periments which demonstrate the significance of poten. We shall show that. \ dt dl/ The new solution differs from the old one just by a phase factor and this corresponds. ih dt ( ih—+i<r. the potentials are needed. a function of function of time. however. as shown in Fig. but short compared with the Suppose we have a charged particle inside a "Faraday length of the tubes. Nevertheless. AHARONOV AND D. The potential in each tube is detercage" connected to an external generator which causes mined by a time delay mechanism in such a way that the potential on the cage to alternate in time. they are In this section. of course. for the region inside the cage. which follows from S=fv(t)dt. BOHM H. In the quantum mechanics. No. we discuss some interesting properties of the electromagnetic potentials in the quantum domain. we shall show that the above conclusions are not correct and that a further interpretation of the potentials is needed in the quantum mechanics. QUANTUM THEORY After the beams pass through the tubes. Now consider a more complex experiment in which a single coherent electron beam is split into two parts and 2. we shall discuss several possible ex.
in which we have a multiply connected region (the region outside the solenoid). We are therefore not surprised that it does not appear in classical mechanics. the beams are brought together at F (Fig. center at the origin and axis in the z direction. can also be expressed as the integral [e/h)^<pdt around a closed circuit in spacetime. but avoiding it. where H = V X A = 0 . evidently.t)=4'iD(x. To demonstrate the effects of the total flux.i)\^2'>(x. we create a magnetic field. The effect is evidently essentially quantummechanical in nature because it comes in the phenomenon of interference.96 486 Y. because the total flux through every circuit containing the origin is equal to a constant <f> Xdx. where <£o= J H < f s = I Xdx. the vector potential. C. where <p is evaluated at the place of the center of the wave packet. BOHM =«£*. in general. As another special case. the problem for each tube is essentially the same as that of the Faraday cage. where \j/i represents the beam on 1 Unless fo^nhc/c. each going on opposite sides of the solenoid. A. let us now consider a p a t h in space only (/= constant). where \f/0 is the solution when A = 0 and where VS/h= (e/c)X.) The beam is split into two parts. But since V is a function only of / wherevdr \j/ is appreciable. FIG. Schematic experiment to demonstrate interference with timedependent scalar potential. we can always obtain a solution for the above Hamiltonian by taking ^/=\l/ve~iSlh. ift: cylindrical metal tubes. (The solenoid can be shielded from the electron beam by a thin plate which casts a shadow. Schematic experiment to demonstrate interference with timeindependent vector potential. The purpose of this arrangement is to ensure that the electron is in a timevarying potential without ever being in a field (because the field does not penetrate far from the edges of the tubes. Thus the potential is nonzero only while the electrons are well inside the tube (region II). From relativistic considerations. B. The phase difference. A. with a coherent beam of electrons. However. Thus. in our problem it is still possible to use such solutions because the wave function splits into two parts ^=^1+^2. F: interference region. The Hamiltonian for this case is ff= It. which is essentially confined within the solenoid.) As in the former example. The solution is then ^ = ^ 1 0 e <Sito__^ 2 0 e iS. H. we begin. it is easily seen that the covariance of the above conclusion demands that there should be similar results involving the vector potential. cannot be zero everywhere outside the solenoid. respectively). in the experiment discussed above. Mi. other edge of the tube./A i suggests that the associated phase shift of the electron wave function ought to be AS/h* chJ where < ^\A<fx=yHds=0 (the total magnetic flux inside the circuit). where n is an integer. il/0e~'slh is a nonsinglevalued function1 and therefore. A. The relativistic generalization of the above integral is _[P(e/c)AJ 2m if(*dt7dx)' where the path of integration now goes over any closed circuit in spacetime. But. 1. When the electron is in region III. 2). (But now there is no need to make wave packets. 2. Wi: wave packets. there is a physical effect of the potentials even though no force is ever actually exerted on the electron. (Si—St)/K. Now let \l'(x. . there is again no potential. T h e above argument In singly connected regions.t) be the wave function when the potential is absent (^i° and ^2° representing the parts that pass through tubes 1 and 2. is evident that the interference of the two parts at F will depend on the phase difference (Si—Sij/h. FIG. Wi. D. tSSUr SSE7 F Electron Beam Metal foil Interference region AHARONOV AND D. and is nonzero only at times when the electron is far from these edges). Nevertheless. By means of a current flowing through a very closely wound cylindrical solenoid of radius R. not a permissible solution of Schr5dinger's equation. This corresponds to another experimental situation. as before. E: suitable devices to separate and divert beams.
444 (1930). But since the main effect of the flux is only to displace the line pattern without changing the interval structure. for example. This corresponds to the setup described in Sec. 377 (1956). alter the sharpness and the general form of the interference bands. Only this time we do not split the plane wave into two parts.6 The solenoid can be used in Marton's device. It would be interesting therefore to test whether such effects actually exist.97 ELECTROMAGNETIC POTENTIALS IN QUANTUM THEORY 487 one side of the solenoid and ^ 2 the beam on the opposite side. a L.8 mm. 5 8 See.) It is easy to devise hypothetical experiments in which the vector potential may influence not only the interference pattern but also the momentum. The electric field would then exist only for a very short time. which will have a discrete set of directions of strong constructive interference. we could do the experiment by observing the interference pattern with and without the magnetic flux. according to our previous discussion. Simpson. L. Such a variation would. but the effects of this field can be made negligible. consider a periodic array of solenoids. Phys. 490 (1953). Alternatively./ *. Each of these beams stays in a simply connected region. rs ! u 1/3 \2 i As yet no direct experiments have been carried out which confirm the effect of potentials where there is no field. For r<R (inside the magnetic field) the solution has been worked out. Rev. (SiSs)/ft= (e/hc) j\dx=(e/hc)<t>t. In order to avoid fully any possible question of contact of the electron with the magnetic field we note that our result would not be changed if we surrounded the solenoid by a potential barrier that reflects the electrons perfectly. Chambers is now making a preliminary experimental study of this question at Bristol. A corresponding shift will take place in the directions. Marton. ^2=^2°e _<s. 1057 (1952). This will be essentially a grating. Z. Consider first the diffraction pattern without the magnetic field. 85. it would be easier to vary the magnetic flux within the same exposure for the detection of the interference patterns. 4 G. 36. there will. In principle. 4. 41 (1955). be an induced electric field outside the solenoid. Rev. This alteration would then constitute a verification of the predicted phenomena. This effect will exist. 4. Phys.) The interference between the two beams will evidently depend on the phase difference. To see this.3 It is quite possible to wind solenoids which are smaller than this. 2 and shown in Fig. an exact solution for this Hamiltonian will be given.3 while the whisker is suitable for another experimental setup4 where the separation is of the order of microns and the whiskers are even smaller than this. 4. when R approaches zero. We therefore can write ^i=^i°e~ iSl/ *. is confirmed in Sec. even though there are no magnetic forces acting in the places where the electron beam passes. 25. of course. Instead.(kr)+bmJ^a)(kr)2. G. We have again chosen the gauge in which Ar=0 and As=<t>/2irr. When the magnetic flux is altered. while the total flux remains fixed. Diiker.2 Recent experiments 34 have succeeded in obtaining interference from electron beams that have been separated in one case by as much as 0. too. Physik 145. 1099 (1954). (In Sec. within the range of present possibilities. each of which is shielded from direct contact with the beam by a small plate. so that only a small part of the beam would be affected by it. suppose the magnetic flux were suddenly altered in the middle of an exposure. where Si and S 2 are equal to (e/c)fkdx along the paths of the first and second beams. The wave equation outside the magnetic field region is. in general of fractional order (dependent on <j>). Sci. this would not be a convenient experiment to do. A PRACTICABLE EXPERIMENT TO TEST FOR THE EFFECTS OF A POTENTIAL WHERE THERE ARE NO FIELDS magnitude has to be of the order of 0o= 2irch/e~4X 10~7 gauss cm2) by means of fine permanently magnetized "whiskers". Mollenstedt. The general solution of the above equation is += £ nH—w —+ +(—+«* I +*•«<). in cylindrical coordinates. Mollenstedt and H. 4. For example. Instr. 2. 90. we may obtain localized lines of flux of the right magnitude (the 2 Dr. The effect of the vector potential will be to produce a shift of the relative phase of the wave function in different elements of the gratings. and therefore the momentum of the diffracted beam. The above solution holds only for r>R. (This. LoV2 r 9r r2\dd / J where k is the wave vector of the incident particle and a = —e4>/ch. Page. (1) eitZamJ^. Acta Met. 3. Such a test is.6 By matching the solutions at r=R it is easily shown that only Bessel functions of positive order will remain. and it will confirm the above results. Marton. in fact. (2) where am and bm are arbitrary constants and Jm+a(kr) is a Bessel function.e. EXACT SOLUTION FOR SCATTERING PROBLEMS We shall now obtain an exact solution for the problem of the scattering of an electron beam by a magnetic field in the limit where the magnetic field region tends to a zero radius. and therefore to place them between the separate beams. Brenner. Sidney S. Naturwissenschaften 42. Rev. respectively. 62 (1956). and Suddeth. .
+1iJc.' * x e f . I t follows t h a t the wave function would not be changed if the electron were kept away from the field b y a barrier whose radius also went to zero with R. I n our cases.E (.*. (8) shall be constant and in the x direction.. a ' . Integraltajel (SpringerVerlag. a 1 (3) So =1 £» 2 m'l (i) ' Jm'+«e ' (ie +i. We shall show in the course of this calculation t h a t the above conditions will be satisfied b y choosing o m = (—j)i m + a l. BOHM This means t h a t the probability of finding the particle inside the magnetic field region approaches zero with R. Of course. where we have used the wellknown formula for Bessel functions: dJy(r)/dr=i(JyiJy+i).>'. so that no problem of multiplevaluedness arises.I sine l —ie*sine } (10) Now rf/! satisfies the simple differential equation d^i oo Because t h e integrand is even in 6. 1949).'+«e i(m ' +I)9 1 — » E {i)m'+a'Jm. let us write it as $\=A\J\—Ii].Ja+iie»Ja]dr'. = E m1 so t h a t ei(o+l)(i»«l) (5) rei«CJit—1») '[. «•». I n the gauge that we are using. 7 See. where we have taken 8 as going from — ir to 7r. \\zx includes Bessel functions of positive order only. we shall have T h e lower limit of the integration is determined by the requirement t h a t when r' goes to zero. . i(i)aeir'°°">. I n order to discuss the asymptotic behavior of ^ i . Hence — = E dr' n. I t is important. 2<a. for example. Hofreiter. ir I t is convenient to split yfr into the following three p a r t s : ^ = ^ 1 + ^ 2 + ^ 3 . 2 sbe I I sine J = E(*')""H" m1 (ID for0>O. d^/dr'= This differential equation can be easily integrated to give j =  —— 2%TH 7KC A*V.) m ' +0. this wave function holds only to the right of the origin. <f/\ also goes to to zero because. (i)maJ^a^imi.**/. Berlin. (9) e '°°<»[./W«ei"<\ f «•>'—[/.+c. Grbbner and N. i gt[aarc B»n(S/fc)] f ^2= E (~i)m+aJm+aeiM.+1 A. as we have seen.e»yr'.(r") +h(i)"Ua+iiei»Jal i cose+i+UV'Va+iiJae"). /8=cos0. we obtain d^i dr' 1 » =•. in which case. k=l.i " (i)m+"J^a'e(^ Jm+al~ Jm+a+1 r*e—•!*'—ie<*l r 1 = e <««'l«l) L =0 r' = kr (6) = ei"»2i" for0<O. T h e general solution in the limit of R tending to zero is therefore «A= E &mJ \nHa\ " e As a result. where 00 72= f T h e first of these integrals is k n o w n 7 : (i)'" +0 <Ai=E /m+„e '»'.e'^™ 2 tn'—o 2m'2 (7) m +a im . AHARONOV AND D. we have written the final expression for the above integral as a function of \8\ and of Isinfll. e**Jtt{kr)=(* 2 /3 2 )* .98 488 Y. W. (4) where +i=AJ A= e"'°°«iJc. We must then choose am so t h a t 41 represents a beam of electrons t h a t is incident from the right (6=0). where 7l= *= E (*)m+0. however. 0<P<k. to satisfy the initial condition t h a t the current density. we easily see t h a t the correct incident wave is ^ i „ c = e .
(2x)»r Now. (*)Jr « i r ' 1 + * * «""' 1~«* . Using now t h e wellknown asymptotic behavior of the error function. 2(27r)»L r . fourth edition.* regard. exrj(iz 2 Vz —* 2 a ' (15) 1 — iexp(—ia 2 ) 2 exp(M )<*z*. "' (asymptotically) only on the upper side of the singuI n \\ "Vfl+rnsfl")2!* larity. > ' dr> (0Jr (*)T «''''' l+«_i* «' r l + « ~ * r e' r ' COs(lTa — ^0) " I / ^ J (2T)»L(rO» ( 0 » cos(Jfl) J (2^')i + « .rI e* e~ir' 'l«»le*l ' (19) 2(2x)»L (/)» l+cos0 (r')»1 . I . it follows that ^ 2 will behave oppositely to 4>i i n this i" .ir).. Jahnke and F. Emde. Inc.. . (12) L (2T)» rv( J. respectively. We obtain ( 9 ). 10 T h e scatcross section is therefore sinVa a= 1 _ (22) ~ E.i v (r') } l+cos0 w(rOMcosflJ1 (r') l+cos0 (/) .ixiaT). The fact that e it is zero for 0<O shows that this part of f i passes ^l_"(~*)'""l">.c o s 0 ) ] * . (16) goes on the upper side of the singularity.. ) M ( _ .(1)" } +i .M « + l ) T . New York. w e find that the term of l / ( r ' ) } in the asymptotic expansion of \pi is 72= f where eir' """(/o+iie»J a )dr'>C+D. To explain this.99 ELECTROMAGNETIC POTENTIALS IN QUANTUM THEORY 489 We shall see presently that Ii represents the largest wefinallyobtain term in the asymptotic expansion of \fn. 9 Reference 8. that our choice of the a„ for Eq. 1943). (3) satisfies the correct boundary conditions. first term in equation (21) represents the incident a n d t h e second t h e scattered wave. 138. p.»L (is) f<° <ir' / 2 \ S C= I e ' > ' ^ [ c o s ( r ' . so that together they will make up the correct \ — — e*r' cos'(—1)6**.)«j ir' e Z)= L ( 2 T ) » [r'(14cos0)2]* h(r'M = Mr'. Tables of Functions (Dover Publications.cosflJ Adding (18) a n d (19) a n d using (11)..W ] () .< we obtain for the corresponding term in $•< dr' / 2 \ d/ /2\i» / Then (rO'Vr/ » gir' coH«rgt[r'l<a+l)«i.) cos(f. p. (20) There remains the contribution of ^3. we find <l/=\l/i+<l/2+t3 —> €«""+*' cosS) H (2«V') J siniracos(0/2) (21) f. in the limit of r' —> 00 we are allowed to take in t h e integrand of I2 t h e first asymptotic term of 7 a . and therefore of the angular ia+l e~"' ] momentum.)ii/ ( / ) _ > ( _ . \nTV ' ' / 2 \ * (i)«+i /•" = 11 I exp(Hz2)rfz \ 7 r / (l+cos0)* •'[r'(i+coH»)i* /2\» »«+» /•» + \ ) M ^ J exp(«Vz. (2ir) [r (l—cos0) J J since according to (5) r(_. 2. (17) — irV'(lcos0) 2 ]»J incident wave. (13) Using again the relation between fi and \j/.— ~ «ir'c<"4. tering t h e ± sign is chosen according t o the sign of a.« r ' « « • « « .e. 8 ?TT rns2f/?/?^ 10 In this way. Similarly. 8 N o w a d d i n 8 ( 1 6 ) a n d ( 17 ) together and using (13) and namely Ja > (2/7rr)» c o s ( r ' . wefinallyg e t (—1)!(" j « . f where The wave. 9 . (12)3 / 2 \ J (14) (_. 24. we verify. whose asymptotic behavior is [see Eq. we note that >pi contains only positive values of m. a . It is quite natural then that this part of ^i H . 2 a Collecting all terms.a). . of course. VT/ ( 1 — COS0)' •'[r'dcorf)]* where we have put z=[r'(lfcos0)]» and z = [ r ' ( l .
Then there would be no trouble in interpreting the fields are. in certain cases. there is no way to reveal this additional richness.—*A„'=A^+d4'/dXil. be conare excluded from it. The first is the case where a = n . (21). (23) transformation.. But it does show that in a theory involving The essential result of the previous discussion is that only local interactions (e. . it is adequate to use the first term of Eq. the fundamental physical entities are the potentials. but. could interact with a field that was a finite distance And since the electrons cannot reach the regions where away. however. according to current nonlocal theory in which. Here. D. which is analogous to that of the index of refration in optics. while the fields are derived from them by differentiations. which is needed on this line. where n is an integer. For this reason. Indeed. the We have seen from the examples described in this behavior of \j/ is not so simple. the essential difference is that the equations of motion of a particle are replaced by the Schrodinger equation for a wave. (It can be seen by direct differentiation that this is a solution.) The second case is that of a=w+§. The main objection that could be raised against the above suggestion is grounded in the gauge invariance of the theory. from the upper or lower side. This Schrodinger equation is obtained from a canonical formalism. the same physical behavior is obtained from any ji »Ir'(l+0OB»))i two potentials. DISCUSSION OF SIGNIFICANCE OF RESULTS the theory. In this case. where ^ is a continuous scalar function. would show that the second term will combine with the first to make a singlevalued wave function. In the interference experiment discussed in Sec. In general.g.100 490 V. all fields must interact only locally. the physical properfields acting on the charged particles. the wave function is \l/==e~ikxe~'"e. which is evidently singlevalued when a is an integer. It was therefore This function vanishes on the line 6=T. which cannot be expressed in terms of the fields alone. The result is In classical mechanics. the single.e. tained for the general case. try to give a further new interpretation to the potenInc.. that its asymptotic behavior is the same as that of Eq.calculate the fields. 2. This confirms the conclusions obtained in the approximate treatment of Sec. The exact solution. for example.Because/ cn+j) W is a closed trigonometric function. This means that insofar as the potenv2 Jo tials are richer in properties than the fields. so that in reality they can be expressed in terms of the directions are clear. 2. we see that the phase of the wave function has a different value depending on whether we approach the line 0 = ± i r from positive or negative angles. Of course. there are severe to the fields themselves.11 a has a maximum when a=n\\. in the neighborhood of 6=w. in this problem. then all the known physical quantities are left unchanged. an electron (for example) can be equation. diffraction effects. Secondly. Af(x) and A/(x). for example. have been neglected. Here. we cannot interpret such effects as due these results. (21) by the scattered wave. in a fieldfree sidered as physically effective. the potentials must. The asymptotic formula (21) holds only when we are not on the line 6=ir. The above discussion suggests that some further It is true that all these effects of the potentials depend development of the theory is needed. if the potentials are subject to the transformation Al. while only the field quantities were thought to have a direct physical meaning. even when there are no multiplyconnected region of space. Bohm. the potentials play a role. as is well known. First. The Lorentz force [eE+(e/c)vXH] does not appear anywhere in the fundamental theory. except insofar as they are used mathematically. but which also requires the potentials. New Jersey. then a vanishes. since \p does not become paper that the above point of view cannot be mainzero on the line 8=TT.. i. related by the above ^. in Schrodinger's equation. we may try to formulate a fields inside the circuit. We shall discuss now the two special cases that can be solved exactly. ties of the system still depend on the potentials. we recall that potentials cannot have such significance because the equation of motion involves only the field quantities themselves. It would therefore seem natural at this point to propose that. A H A R O N O V A N D D. In quantum mechanics. the integrals for ^ can be carried out exactly. However. Englewood Cliffs. 1951). difficulties in the way of doing this. Schrodinger's or Dirac's in quantum theory. the electron relativistic notions. Quantum Theory (PrenticeHall.=_ei(i»+r'co8») / exp(is2)<fe. represented in Eq. It can be seen concluded that the potentials cannot have any meaning. BOHM When a=n. valuedness of ^ is evident. we may retain the present local theory and. Therefore. Two possible only on the gaugeinvariant quantity jfkdx= fHds. In other words. to (2) with a set equal to « + § . As a result. This is analogous to the Ramsauer effect. but appears only as an approximation holding in the classical limit. we may 11 See. instead. We shall see this in more detail presently for the special case a=«+§. our discussion does not bring into question the gauge invariance of 5. In other words. in quantum mechanics. despite the nonsinglevalued character of the two parts. the potentials have been regarded as purely mathematical auxiliaries. and current quantummechanical field theinfluenced by the potentials even if all the field regions ories).
Ltd. ACKNOWLEDGMENTS We are indebted to Professor M. Vol. Such states have actually been used in connection with recent theories of superconductivity and superfluidity12 and we shall show their relation to this problem in more detail.. H. edited by N. Pryce for many helpful discussions. p. . Sec. Advances in Physics.101 ELECTROMAGNETIC POTENTIALS IN QUANTUM THEORY 491 tials. L. 8. In other words. 3. for example. u See. can be introduced. 25. F. G. It will be shown in a future paper that in a system containing an undefined number of charged particles (i. which is conjugate to the charge density and which may give a meaning to the gauge.. This means that we must be able to define the physical difference between two quantum states which differ only by gauge transformation. essentially an angle variable. Chambers for many discussions connected with the experimental side of the problem. we are led to regard Ai. a new Hermitian operator. C. London. We wish to thank Dr. a superposition of states of different total charge). Mott (Taylor and Francis. 3. Par.e. 1959). Kuper.(x) as a physical variable.
observing plane o.2 To obtain a more direct check. 1) are about 6. This explains how Marton et^al. 1960) Aharonov and Bohm1 have recently drawn attention to a remarkable prediction from quantum theory. 1. 1) flanked by two earthed metal plates e. e. According to this.102 SHIFT OF AN ELECTRON INTERFERENCE PATTERN BY ENCLOSED MAGNETIC FLUX R. A straightforward calculation shows that in a "biprism" experiment. 1). 4 such a field should produce a fringe displacement which exactly keeps pace with the deflection of the beams by the field.g.5 fi in diameter and a 3 FKJ. experimentally. It has since been pointed out2 that the same conclusion had previously been reached by Ehrenberg and Slday. a Philips EMI00 electron microscope 8 has been modified so that it can be switched at will from normal operation to operation as an interferometer.o (Fig. but these authors perhaps did not sufficiently s t r e s s the remarkable nature of the result. With this microscope it was not possible to reduce the virtual source diameter below about 0.2 jx. and p r o duce a fringe shift within it. and a field of type a' should leave the fringes undisplaced in space. 1. Is the effect on the fringe system of Stray fields not localized to region a but extending. It Is convenient to refer to a natural "flux unit. respectively.g. with •owee s. even though the beams themselves pass only through fieldfree regions. and their work appears to have attracted little attention. there will then be a shift due simply to curvature of the electron trajectories by the field. so that it was necessary to use a fiber f only about 1. Theory predicts a shift of n fringes for an enclosed flux * of nhc/e. Clearly the first problem to consider." hc/e =4. England (Received May 27. on the other hand.7 cm and 13.. 4 The distances sf a n d / . Schematic diagram of interferometer.. H.5 were able to observe fringes in the presence of stray 60cps fields probably large enough to have destroyed them otherwise. the fringe pattern In an electron interference experiment should be shifted by altering the amount of magnetic flux passing between the two beams (e. and confined and extended Held regions a and a'. should leave the envelope undisplaced. in region a of Fig. this experiment thus constitutes an inadvertent check of the existence of the "quantum" shift. In addition to the "quantum" fringe shift due to the enclosed flux. G. In the Marton 5 interferometer. A field of type a.' using semiclassical arguments. over region a' in Fig. Fringes are produced by an electrostatic "biprism" consisting of an aluminized quartz fiber / (Fig. so that the fringe system appears to remain undisplaced relative to the envelope of the pattern.4 cm. conditions are different. Chambers H. / . . Bristol. Wills Physios Laboratory. biprism e. altering the positive potential applied to / alters the effective angle of the biprism.135 xlO" 7 gauss cm'. University of Bristol.
1). The biprism fringes are now unperturbed. We also verified that with this interferometer. . from a different part of the same whisker. the flux content will change along the length at a rate d$/dz of about 1 flux unit per 4 micron. but the fringe system within the envelope is inclined at an angle of the order of one fringe width per micron. the fringe width in the observing plane o is about 0. in diameter and 0. just outside shadow of fiber. These results confirm the presence of the quantum shift in fields of type a'. we expect to see a pattern in which the envelope is undisplaced. Precisely this is observed experimentally. 3. 3(a) the whisker (a) (b) (c) FIG. and with fiber out of the field of view. shows the pattern in a field producing a displacement of about 2. Such a field was produced by an iron whisker.103 VOLUME 5. where intuition might expect no effect. (a) Fringe pattern due to biprism alone. which is extremely convenient for the present purpose. placed in the shadow of the fiber / . 1960 IK \\ (a) (b) FIG.6 LI.6 jx. and there is a "pinhole" magnification of x3 between the biprismfiber assembly and the observing plane. Figure 3(c) shows a further example of these fringes. 3(b) shows the same section of whisker as Fig. Figure 2(b). Fields up to 0.5 fringe widths. We first examined the effect of a field of type a'. but in this example becomes very small in the upper part of the picture. These fringe shifts cannot be attributed to direct interaction between the electrons and the surface of the whisker. It will be seen that the whisker taper is not uniform. Thus if such a whisker is placed in position a (Fig. (b) Fresnel fringes in the shadow of the whisker itself. very small biprism angle. we thus expect the fringes to show a tilt of order 1 in 5 relative to the envelope of the pattern. this pattern would have had the light and dark fringes interchanged. 7 about 1 JJ. as shown in Fig. moved just out of the shadow of the biprism fiber. 5 fringe widths by field of type a'. Of more interest is the effect predicted for a field of type a. produced by a Helmholtz pair of single turns 3 mm in diameter just behind the biprism. with the biprism moved out of the way. unlike Marton's. Since the fringe width in the observing plane is 0. An iron whisker 1 LI in diameter will contain about 400 flux units. a small ac field suffices to blur out the fringe system completely. to produce a wide pattern of fringes which would not be blurred out by the finite source size. (c) Same as (b).5 mm long. the Fresnel fringes in the shadow of the whisker show exactly the same pattern of fringe shifts along their length as in Fig.3 gauss were applied. In fact the biprism is an unnecessary refinement for this experiment: Fresnel diffraction into the shadow of the whisker is strong enough to produce a clear fringe pattern from the whisker alone. In the absence of the "quantum" shift due to the enclosed flux. 3(a). moreover they are found to taper 9 with a slope of the order of 10~3. 3(a). 2. (a) Tilted fringes produced by tapering whisker in shadow of biprism fiber. but from a different part of the whisker. 2(a). The fringe pattern obtained is shown in Fig. NUMBER 1 PHYSICAL REVIEW LETTERS JULY 1. (b) Pattern displaced by 2. if it tapers uniformly with a slope of 10" 3 . 3(a). Thus Fig. and as predicted the appearance of the pattern was completely unchanged. sufficient to displace the pattern by up to 30 fringe widths. The whisker here is tapering more rapidly. Whiskers as thin as this are expected theoretically 8 and found experimentally 9 to be single magnetic domains. for instance. since in Fig.
115. the different •Igns applying to paths on either side of the whisker. Phys. 25. due to the flux emerging from the sides and ends. 3 W. 62 (1956). Simpson. Nor can they be attributed to a return field Hz parallel to the whisker in the region a' outside it. W. Ehrenberg and R. for two reasons. Aharonov and Dr. thus skewing the interference fringes. 3 might be taken to demonstrate the existence of the predicted quantum shift. DeBlois. I am indebted to Mr. Sci. 70. and J. 6 A. The two beams which converge to Interfere at o a r e thus tilted one above and one below the plane of Fig. 5 L. Physik 145. and amounts to a phasedifference gradient of («/*c)d*/dz. 485 (1959). 1099 (1954). 4 G. There is a progressive change In the phase difference between the two beams at one moves in the z direction. Proc. that the total displacement of a given fringe is a direct measure not of the leakage field from that section but of the flux enclosed within it. 965 (1946). 4. Bohm. Soc. Kittel. 377 (1956). Phys. Rev. 1960 la completely in the shadow of the fiber. *Y. and it is illuminating to consider this.104 5. Rev. and A. Suddeth. One thus sees fairly intuitively how the "quantum" phase difference is progressively built up from the free end of the whisker. Rev. van Dorsten. C. A. H. 344 (1960). W. This Is precisely the rate of change Of the "quantum" phase difference e$/Kc calcu lated by Aharonov and Bohm. 8 C. Marton. Aharonov and D. 1. Instr. Z. where it is zero. 10 Immediately outside a tapering whisker. Phys. to any section where the interference is being observed. J. we have seen experimentally that an extended field Hz in fact produces a completely different effect (Fig. and observation confirms that the displacement of the envelope is very small. 12. J. G. Werner and D. Brill. Philips Tech. 2 F. 8(1949). Diiker. 33 (1950). It remains true. 9 R. Nieuwdorp. Indeed they do. R. 7 The use of a whisker was first suggested by Dr. Verhoeff. Acta Met. An estimate of the magnitude of this field shows that it might be strong enough to displace the pattern by perhaps one fringe width. Mitchell. nevertheless the tilt of the fringes can be attributed to a leakage field. . Rev. Rev. Letters 4. 10 The following analysis Is due to Professor Pryce. Appl. J. 459 (1958). This (laid exerts a force on the electron and gives it a momentum pz =± \(e/c)d$/dz. (London) B62. the leakage field is in fact primarily radial and Is given by Hr = (d$/dz)/2nr. This is easily calculated from pz by de Broglie's relation. NUMBER 1 PHYSICAL RE IEW LETTERS JULY 1. 29. Siday. as Pryce has pointed out to me. and that a displacement will occur even in a parallelsided region of the whisker where the radial leakage field is zero. Mollenstedt and H. Bohm for telling me of their work before publication. E. The whiskers were grown by the method of 5. A. and to them and to Professor Pryce for many discussions. S. however. Brenner. Phys. Thus the patterns of Fig. 2). but not more. Phys. secondly.
105
PHYSICAL REVIEW VOLUME 96, NUMBER 1 OCTOBER 1, 1954
Conservation of Isotopic Spin and Isotopic Gauge Invariance*
C. N. YANG t AND R. L. MILLS
Brookhaven National Laboratory, Upton, New York (Received June 28, 1954) It is pointed out that the usual principle of invariance under isotopic spin rotation is not consistant with the concept of localized fields. The possibility is explored of having invariance under local isotopic spin rotations. This leads to formulating a principle of isotopic gauge invariance and the existence of a b field which has the same relation to the isotopic spin that the electromagnetic field has to the electric charge. The b field satisfies nonlinear differential equations. The quanta of the b field are particles with spin unity, isotopic spin unity, and electric charge ± e or zero. INTRODUCTION stable even nuclei contain equal numbers of them. Then in 1937 Breit, Condon, and Present pointed out the approximate equality of p—p and n—p interactions in the ! 5 state. 2 I t seemed natural to assume t h a t this equality holds also in the other states available to both the n — p and p — p systems. Under such an assumption one arrives a t the concept of a total isotopic spin 3 which is conserved in nucleonnucleon interactions. Experi2 Breit, Condon, and Present, Phys. Rev. 50, 825 (1936). J. Schwinger pointed out that the small difference may be attributed to magnetic interactions [Phys. Rev. 78, 135 (1950)]. 3 The total isotopic spin T was first introduced by E. Wigner, Phys. Rev. 51, 106 (1937); B. Cassen and E. U. Condon, Phvs. Rev. 50, 846 (1936).
T
H E conservation of isotopic spin is a much discussed concept in recent years. Historically an isotopic spin parameter was first introduced b y Heisenberg 1 in 1932 to describe the two charge states (namely neutron and proton) of a nucleon. T h e idea that the neutron and proton correspond to two states of the same particle was suggested a t t h a t time by the fact that their masses are nearly equal, and t h a t the light
W ork performed under the auspices of the U. S. Atomic Energy Commission. t On leave of absence from the Institute for Advanced Studv, Princeton, New Jersey. 1 W. Heisenberg, Z. Physik 77, 1 (1932).
106
192 C. N. VANG AN D R . L. MILLS
ments in recent years4 on the energy levels of light nuclei strongly suggest that this assumption is indeed correct, An implication of this is that all strong interactions such as the pionnucleon interaction, must also satisfy the same conservation law. This and the knowledge that there are three charge states of the pion, and that pions can be coupled to the nucleon field singly, lead to the conclusion that pions have isotopic spin unity. A direct verification of this conclusion was found in the experiment of Hildebrand5 which compares the differential cross section of the process n+p—^+d with that of the previously measured process p+p^m*\d. The conservation of isotopic spin is identical with the requirement of invariance of all interactions under isotopic spin rotation. This means that when electromagnetic interactions can be neglected, as we shall hereafter assume to be the case, the orientation of the isotopic spin is of no physical significance. The differentiation between a neutron and a proton is then a purely arbitrary process. As usually conceived, however, this arbitrariness is subject to the following limitation: once one chooses what to call a proton, what a neutron, at one spacetime point, one is then not free to make any choices at other spacetime points. It seems that this is not consistent with the localized field concept that underlies the usual physical theories. In the present paper we wish to explore the possibility of requiring all interactions to be invariant under independent rotations of the isotopic spin at all spacetime points, so that the relative orientation of the isotopic spin at two spacetime points becomes a physically meaningless quantity (the electromagnetic field being neglected). We wish to point out that an entirely similar situation arises with respect to the ordinary gauge invariance of a charged field which is described by a complex wave function ip. A change of gauge6 means a change of phase factor ip—np', 4/>= (expia)^, a change that is devoid of any physical consequences. Since ^ may depend on x, y, z, and /, the relative phase factor of ^ at two different spacetime points is therefore completely arbitrary. In other words, the arbitrariness in choosing the phase factor is local in character. We define isotopic gauge as an arbitrary way of choosing the orientation of the isotopic spin axes at all spacetime points, in analogy with the electromagnetic gauge which represents an arbitrary way of choosing the complex phase factor of a charged field at all spacetime points. We then propose that all physical processes (not involving the electromagnetic field) be invariant under an isotopic gauge transformation, ip—rf't ^'=S~^, where S represents a spacetime dependent isotopic spin rotation. To preserve invariance one notices that in electro*T. Lauritsen, Ann. Rev. Nuclear Sci. 1, 67 (1952); D. R. Inglis, Revs. Modern Phys. 25, 390 (19S3). 6 R. H. Hildebrand, Phys. Rev. 89, 1090 (1953). 6 W. Pauli, Revs. Modern Phys. 13, 203 (1941).
dynamics it is necessary to counteract the variation of a with x, y, z, and t by introducing the electromagnetic field A,, which changes under a gauge transformation as
A;=A>+—.
1 da
e 6% In an entirely similar manner we introduce a B field in the case of the isotopic gauge transformation to counteract the dependence of S on x, y, z, and t. It will be seen that this natural generalization allows for very little arbitrariness. The field equations satisfied by the twelve independent components of the B field, which we shall call the b field, and their interaction with any field having an isotopic spin are essentially fixed, in much the same way that the free electromagnetic field and its interaction with charged fields are essentially determined by the requirement of gauge invariance. In the following two sections we put down the mathematical formulation of the idea of isotopic gauge invariance discussed above. We then proceed to the quantization of the field equations for the b field. In the last section the properties of the quanta of the b field are discussed. ISOTOPIC GAUGE TRANSFORMATION Let f be a twocomponent wave function describing a field with isotopic spin J. Under an isotopic gauge transformation it transforms by
+=sy,
(i)
where 5 is a 2X2 unitary matrix with determinant unity. In accordance with the discussion in the previous section, we require, in analogy with the electromagnetic case, that all derivatives of $ appear in the following combination: (3^—ieB^ip. B,, are 2X2 matrices such that 7 for n= 1, 2, and 3, B^ is Hermitian and 5 4 is antiHermitian. Invariance requires that S(dllitBll'W=(dllieBli)$. (2)
Combining (1) and (2), we obtain the isotopic gauge transformation on B?: i B^S'B.S+S1—. e dS (3) 6%
The last term is similar to the gradiant term in the gauge transformation of electromagnetic potentials. In analogy to the procedure of obtaining gauge invariant field strengths in the electromagnetic case, we
' W e use the conventions fi=c=l, and xt—it. Boldface type refers to vectors in isotopic space, not in spacetime.
107 193
ISOTOPIC
SPIN
AND ISOTOPIC
GAUGE
INVARIANCE
define now dBp dB„ F„= dxv dx,, One easily shows from (3) that (5) under an isotopic gauge transformation.! Other simple functions of B than (4) do not lead to such a simple transformation property. The above lines of thought can be applied to any field 41 with arbitrary isotopic spin. One need only use other representations 5 of rotations in threedimensional space. I t is reasonable to assume that different fields with the same total isotopic spin, hence belonging to the same representation S, interact with the same matrix field Bp. (This is analogous to the fact that the electromagnetic field interacts in the same way with any charged particle, regardless of the nature of the particle. If different fields interact with different and independent B fields, there would be more conservation laws than simply the conservation of total isotopic spin.) To find a more explicit form for the B fields and to relate the B/s corresponding to different representations S, we proceed as follows. Equation (3) is valid for any 5 and its corresponding Bf. Now the matrix S~ldS/dxp appearing in (3) is a linear combmation of the isotopic spin "angular momentum" matrices T* (*=1, 2, 3) corresponding to the isotopic spin of the \j/ field we are considering. So B? itself must also contain a linear combination of the matrices T\ But any part of B? in addition to this, B^, say, is a scalar or tensor combination of the T's, and must transform by the homogeneous part of (3), BJ = S^B^S. Such a field is extraneous; it was allowed by the very general form we assumed for the Bfield,but is irrelevant to the question of isotopic gauge. Thus the relevant part of the B field is of the form
jBj,= 2b u T.
HtiflrB.BtB,).
(4)
But the sum of B^ and B„w, the B fields corresponding to SM and 5 ( w , transforms in exactly the same way, so that
B^B^+B^
(plus possible terms which transform homogeneously, and hence are irrelevant and will not be included). Decomposing 5 ( o ) 5 t w into irreducible representations, we see that the twelvecomponent field b,, in Eq. (6) is the same for all representations. To obtain the interaction between any field ^ of arbitrary isotopic spin with the b field one therefore simply replaces the gradiant of f by (d„2«b„T>/',
i
(7)
where T (i= 1, 2, 3), as defined above, are the isotopic spin "angular momentum" matrices for thefield\p. We remark that the nine components of b,,, /*= 1, 2, 3 are real and the three of b 4 are pure imaginary. The isotopicgauge covariant field quantities F^ are expressible in terms of bM: F„=2U,T, where dbp 3b „ (8)
(9)
t,
dx, dx,.
2eb,,Xb„
ip, transforms like a vector under an isotopic gauge transformation. Obviously the same fM„ interact with all fields ij/ irrespective of the representation S that ^ belongs to. The corresponding transformation of b^ is cumbersome. One need, however, study only the infinitesimal isotopic gauge transformations, Then 5=1—2iT5io. 1 d b / = b,+2bMX«o>H
FIELD EQUATIONS
5<d.
t dXp
(10)
(6)
(Boldface letters denote threecomponent vectors in isotopic space.) To relate the b / s corresponding to different representations 51 we now consider the product representation 5 = 5 ( o ) 5 ( W . The B field for the combination transforms, according to (3), by
To write down the field equations for the bfieldwe clearly only want to use isotopic gauge invariant quantities. In analogy with the electromagnetic case we therefore write down the following Lagrangian density :8 —if .f Since the inclusion of a field with isotopic spin \ is illustrative, and does not complicate matters very much, we shall use the following total Lagrangian density: £=—if^vfM»—1^7^(3^—itzb^ip—m^l'. (11)
w
i
+ _[ 5 («)]i
as<<"
dxu
i
1
ds< w
e
([St"]. e dxu
t Note added in proof.—It may appear that B„ could be introduced as an auxiliary quantity to accomplish invariance, but need not be regarded as afieldvariable by itself. It is to be emphasized that such a procedure violates the principle of invariance. Every quantity that is not a pure numeral (like 2, or M, or any definite representation of the y matrices) should be regarded as a dynamical variable, and should be varied in the Lagrangian to yield an equation of motion. Thus the quantities Bp must be regarded as independent fields.
One obtains from this the following equations of motion: 3Wa*„+2e(b,,Xf,.„)+J„=0,
(12)
7n(d„—iez • b„)\l'+m\l/ = 0,
8 Repeated indices are summed over, except where explicitly stated otherwise. Latin indices are summed from 1 to 3, Greek ones from 1 to 4.
108 194 where
J)J = iei/'7 ( ,Ti/<.
YANG
AND
R .
M ILLS
(13)
The divergence of JF does not vanish. Instead it can easily be shown from (13) that diJdxll=2t\Xi,If we define, however, 3 . = ^+2£b„Xf,„, then (12) leads to the equation of continuity, a3'p/dz„ = 0. (16) (15) (14)
This is the analog of the equation da/dx/. = 0 that must be satisfied by the gauge transformation A J = Al2+e~1(da/dxll) of the electromagnetic field.
QUANTIZATION
To quantize, it is not convenient to use the isotopic gauge invariant Lagrangian density (11). This is quite similar to the corresponding situation in electrodynamics and we adopt the customary procedure of using a Lagrangian density which is not obviously gauge invariant : £= 3b„ 2t(b„Xb„)— 2 dx„ 3.r, d.r„ e 2 (b,Xb,) 2 +J„b,^(7,a,+OT)^. (19) The equations of motion that result from this Lagrangian density can be easily shown to imply that
a2
1 db,, 3bM
31,2,3 and 34 ar.e respectively the isotopic spin current density and isotopic spin density of the system. The equation of continuity guarantees that the total isotopic spin l = i ^4a°x
• /
is independent of time and independent of a Lorentz transformation. It is important to notice that 3tM, like b„, does not transform exactly like vectors under isotopic space rotations. But the total isotopic spin, din c ~d?x,
dXi
d
where
a42eb„X a = 0, dx? dxv a = dbn/'dXp.

/
is the integral of the divergence of f , , which transforms 4like a true vector under isotopic spin space rotations. Hence, under a general isotopic gauge transformation, if 5—>Sn on an infinitely large sphere, T would transform like an isotopic spin vector. Equation (15) shows that the isotopic spin arises both from the spin! field (J,,) and from the b,, field itself. Inasmuch as the isotopic spin is the source of the b field, this fact makes the field equations for the b field nonlinear, even in the absence of the spinJ field. This is different from the case of the electromagnetic field, which is itself chargeless, and consequently satisfies linear equations in the absence of a charged field. The Hamiltonian derived from (11) is easily demonstrated to be positive definite in the absence of the field of isoiopi" spin \. The demonstration is completely identical with the similar one in electrodynamics. We must complete the'set of equations of motion (12) and (13) by the supplementary condition, db„/dx„=0, (17) which serves to eliminate the scalar part of the field in b^. This clearly imposes a condition on the possible isotopic gauge transformations. That is, the infinitesimal isotopic gauge transformation 5=1—izba> must satisfy the following condition: d 1 d2 2b M X—5GH 5u=0. dx„ e dxJ (18)
Thus if, consistent with (17), we put on one spacelike surface a = 0 together with da/dt = 0, it follows that a = 0 at all times. Using this supplementary condition one can easily prove that the field equations resulting from the Lagrangian densities (19) and (11) are identical. One can follow the canonical method of quantization with the Lagrangian density (19). Defining n„=3bM/to4+2e(b„Xb4), one obtains the equaltime commutation rule [ V M , n„'(*')]<=('=  5 0  M 3 ( *  * ' ) , (20)
where bj, i= 1, 2, 3, are the three components of b^. The relativistic invariance of these commutation rules follows from the general proof for canonical methods of quantization given by Heisenberg and Pauli.9 The Hamiltonian derived from (19) is identical with the one from (11), in virtue of the supplementary condition. Its density is
1 3b„ db?
ff0= £n„ n„+
2 dXj dXj
+HyA+m)f,
(21)
Hint=2e(biXbA)ni2t(bl,Xbj)(dbll/dx,) + e 2 (b i Xb ; ) 2 J„b„, The quantized form of the supplementary condition is the same as in quantum electrodynamics. ' W. Heisenberg and W. Pauli, Z. Physik 56, 1 (1929).
109
ISOTOPIC
SPIN
AND
ISOTOPIC
GAUGE
IN V A R I A N C E
195
PROPERTIES OF T H E b QUANTA
The quanta of the b field clearly have spin unity and isotopic spin unity. We know their electric charge too because all the interactions that we proposed must satisfy the law of conservation of electric charge, which is exact. The two states of the nucleon, namely proton and neutron, differ by charge unity. Since they can transform into each other through the emission or absorption of a b quantum, the latter must have three charge states with charges ± e and 0. Any measurement of electric charges of course involves the electromagnetic field, which necessarily introduces a preferential direction in isotopic space at all spacetime points. Choosing the isotopic gauge such that this preferential direction is along the z axis in isotopic space, one sees that for the nucleons Q= electric chs.vge= e(^+e_1T'), and for the b quanta Q=(e/t)T: The interaction (7) then fixes the electric charge up to an additive constant for all fields with any isotopic spin: Q=e(e>T*+R). (22) The constants R for two charge conjugate fields must be equal but have opposite signs.10
F I G . 1. Elementary vertices for b fields and nucleon fields. Dotted lines refer to b field, solid lines with arrow refer to nucleon field. I T J""" * [ i \ / / V / \
'
l \
FIG. 2. Primitive divergences.
(
I
d
e
satisfies
idi>/dt = Hir,ti>,
/' * N
V
x
We next come to the question of the mass of the b quantum, to which we do not have a satisfactory answer. One may argue that without a nucleon field the Lagrangian would contain no quantity of the dimension of a mass, and that therefore the mass of the b quantum in such a case is zero. This argument is however subject to the criticism that, like all field theories, the b field is beset with divergences, and dimensional arguments are not satisfactory. One may of course try to apply to the b field the methods for handling infinities developed for quantum electrodynamics. Dyson's approach11 is best suited for the present case. One first transforms into the interaction representation in which the state vector ^
10 11
where Hint was defined in Eq. (21). The matrix elements of the scattering matrix are then formulated in terms of contributions from Feynman diagrams. These diagrams have three elementary types of vertices illustrated in Fig. 1, instead of only one type as in quantum electrodynamics. The "primitive divergences" are still finite in number and are listed in Fig. 2. Of these, the one labeled a is the one that effects the propagation function of the b quantum, and whose singularity determines the mass of the b quantum. In electrodynamics, by the requirement of electric charge conservation,12 it is argued that the mass of the photon vanishes. Corresponding arguments in the b field case do not exist13 even though the conservation of isotopic spin still holds. We have therefore not been able to conclude anything about the mass of the b quantum. A conclusion about the mass of the b quantum is of course very important in deciding whether the proposal of the existence of the b field is consistent with experimental information. For example, it is inconsistent with present experiments to have their mass less than that of the pions, because among other reasons they would then be created abundantly at high energies and the charged ones should live long enough to be seen. If they have a mass greater than that of the pions, on the other hand, they would have a short lifetime (say, less than 1020 sec) for decay into pions and photons and would so far have escaped detection.
'}. Schwinger, Phys. Rev. 76, 790 (1949). 13 In electrodynamics one can formally prove that G„,,£„ = 0, where GfU is defined by Schwinger's Eq. (A12). {G^A, is the current generated through virtual processes by the arbitrary external field .1„.) No corresponding proof has been found for the present case. This is due to the fact that in electrodynamics the conservation of charge is a consequence of the equation of motion of the electron field alone, quite independently of the electromagnetic field itself. In the present case the b field carries an isotopic spin and destroys such general conservation laws.
See M . GellMann, Phys. Rev. 92, 833 (1953). F. J. Dyson, Phys. Rev. 75, 486, 1736 (1949).
The problem of particle types and other contributions to the theory of elementary particles
Ronald Shaw Cambridge University Ph.D. thesis, September 1955
1 2
2.1 2.2 2.3
2.3.1
Representations of the improper Lorentz group and the problem of particle types Some contributions to the theory of elementary particles
Some results in the theory of interacting fields The associated production of 0° and A0 particles Invariance under general isotopic spin transformations
Introduction
Many general results follow from postulating that the Lagrangian of the system of fields should be invariant under certain classes of transformations. For example, if the Lagrangian is invariant under translations and rotations in spacetime, then it is well known that a momentum 4vector Pa and an angular momentum tensor Map exist and are conserved. These two invariances seem to exhaust the useful possibilities for transformations in spacetime. However one can also postulate the invariance of the Lagrangian under rotations in a Euclidean space En of dimension n which has no relation to spacetime, and useful results still follow.
The invariance of the Lagrangian under special gauge transformations can be looked upon from this point of view. The space involved is then E2, and invariance under rotations in this space results in the existence of a divergenceless current vector sa and of a charge Q which is conserved. Neutral fields are scalars in E2, while fields with charge one are vectors in Ei For instance, taking the case of two real fermion fields, tpi and 02 with free Lagrangian 1
L
o = i{i)lB~jadai>i
 daiJ)iB"jail)i) + im^Bipi
(1)
it is seen that Lo is invariant under the infinitesimal rotation i>\ = i'i  c4>2 4^2 = 4>2 + Ci'l
(2)
(In equation (1) B is the matrix defined by B~jaB~l — — 7J; it then follows that ipBxp is a scalar, ipB^if) a vector, etc.) The current density and total charge are in this case sa = ie{i\)XB^a^2  ^Bjaipi) (3)
Q = jSadaa
(4)
In the special gauge transformations (2), c was a constant independent of the spacetime argument of ij>. If now the Lagrangian is required to be invariant under general gauge transformations (i.e. those with c a function of position) then it is found necessary to introduce the electromagnetic field. This method was adopted by Schwinger (1954). For under general gauge transformations, Lo is no longer invariant but becomes Lo' = L0idaC^B^fa  0 2 L?7 a V'i) (5)
However the last term of (5) can be compensated by taking an interaction Lagrangian I i = Aasa (6) and requiring that when ip undergoes the transformation (2), Aa should undergo the transformation eA'a = eAa + dac (7)
Lorentz suffices are represented by Greek letter, and suffices in E„ by Latin letters. The summation convention is employed with respect to both kinds of suffices.
1
Finally, defining as usual the field strengths fap = daA/3  dpAa and a free Lagrangian L2 for the electromagnetic field (8)
L2 =
the total Lagrangian
\urp
0)
(10)
L = L0 + Li + L2
is seen to be invariant under general gauge transformations. The purpose of the above treatment of the transition from special gauge transformations to general gauge transformations was to present the introduction of the electromagnetic field in a form suitable for generalization to spaces En with n > 2. For in the usual symmetrical meson theory (Kemmer, 1938), invariance under rotations in a space £ 3 , the socalled isotopic spin space, is assumed. An infinitesimal rotation in this space involves three real parameters c, instead of the one parameter c of the transformation (2). In the usual theory the c, do not depend on position, or in other words, only special isotopic spin transformations are contemplated. However, in analogy with the case of E2, it is natural to postulate invariance of the Lagrangian under general isotopic spin transformations, the c, being functions of position. The consequences 2 of the postulate are worked out in Section 2.3.2. In Section 2.3.3, the SalamPolkinghorne scheme, in which all the strong interactions are assumed to be invariant under rotations in a space E4, is generalised in a similar manner. 2.3.2 T h e invariance in the space E3
The fermion case is taken once more, with free field Lagrangian L0 = 4>(iada + m)4' (11)
where i\) is a 4component spinor in spacetime and a 2component spinor in E3. Note that this is in contrast to the case of E2l when i\) was taken to be
The work described in this chapter was completed, except for its extension in Section 2.3.3, in January 1954, but was not published. In October 1954, Yang and Mills adopted independently the same postulate and derived similar consequences.
2
In analogy with (3) and (4) the isotopic spin current density sa and the total isotopic spin T can be defined Sa = + .In (11) the two isotopic spin components of ip refer of course to the proton and neutron. Under a rotation </>i = <j>i + Cirfj in isotopic spin space. — * (15) sa is seen to be divergenceless and hence T is conserved. transforms qB'a = q(Ba . (18) In order to construct a theory invariant under general isotopic spin transformations it only remains to construct an invariant free field Lagrangian for the Bfield. are the usual isotopic spin matrices. In (14) q is the unit of 'isotopic charge' and corresponds to e in (3). = \tijkCjk a n d where T.1 = ( l + UCIT^ V (13) (12) where c.a vector in E2.' i # 7 a f ^ and (14) f=lsadaa. ip undergoes the transformation 4. L0 is no longer invariant but becomes L'0 = U + \idac^lar^.F (17) (16) and by requiring that when ^ transforms under (13) the field Ba. which is the generalization to E3 of the electromagnetic field. in analogy with the electromagnetic case. The additional term can be compensated by introducing an interaction Lx = Ba.e x Ba) + dac. It is natural. Now under general transformations (13) with c a function of position. to define .
as Fa0 transforms solely as a vector in E3.* (20) However (20) is not invariant under the transformation (18). follows immediately from (26) j a = s" .FaP (23) x Ba) (22) instead of (20).q(Ba x Bp Bp and by taking L2 to be L2 = l~F^. A properly invariant Lagrangian can be obtained by defining Fap = faP . (28) . For f°^ transforms as follows jtap = f°P_gx pP _ 2(dac x B " .dpc x Ba) (21) The second order derivatives of c have cancelled as in the electromagnetic case.and to take the Lagrangian to be L2 = \faPfap — . The filed equations which follow from the Lagrangian L = L0 + LX + L2 are found to be hada + m)ip = and dpFap + 2qBp x Fa(3 = s° (26) The divergence of s" no longer vanishes. the total isotopic spin is now (27) l (24) iqBa. L2 is now invariant under (18). but a divergenceless vector j a .2qBp x FaP and instead of (15). say.rlc4 (25) f = jfdaa. but the first order derivatives still remain and spoil the invariance of (20).
i * 3 . The fact that the Sfield itself carries isotopic spin accounts for the greater complexity of equation (26) compared with the corresponding equation —dpfap = sa in the electromagnetic case. The resulting equations are slightly more cumbersome as vector notation cannot be employed.:(7ad a + m)tf. and is composed of the isotopic spin s" of the nucleons and the isotopic spin —2q£r x Faf3 of the Bfield. The status of invariances like this. like the 7rmeson.2.(30) where \P. It is clear that these particles. is obscure and is to be contrasted with invariance under gauge transformations. that of electric charge. seems never to be violated. 2. but the conclusion is that gauge invariance and isotopic spin invariance do not appear to differ essentially as far as generalization to spacedependent transformations is concerned. if they exist.= *o . Zero mass would appear to be ruled out since otherwise the neutron would have a rapid decay N>P + B~ (29) In fact one would have expected the Bparticles to have been observed unless their mass was quite large or unless the coupling constant q was very small.is a 4vector in E4 and describes both nucleons and cascade particles. 22 XEo = * 1 + *'*2 J . where the ensuing conservation law.j a replaces s" as the isotopic spin current density of the system. will have spin 1 and isotopic spin 1.. and it is not clear from (26) whether the mass of the Sfield is zero or not.3. and so exist in three states with charge ± e and 0. The connection between <I>.3 Invariance in the space E4 In this section the 4dimensional theory of Salam and Polkinghorne (1955) is generalised.3. which apply only to part of the Lagrangian. The existence of this contrast was the motive for the above work. This means that it is difficult to determine whether or not the existence of the Bfield is consistent with experiment. The fermion free field Lagrangian is now U = *. Finally it should be pointed out that the the invariance postulated in this section does not hold when the electromagnetic field is taken into account. in a way similar to the 3dimensional isotopic spin case of section 2. and these fields can be taken as 2 » x s .
'J transforms *'< = % + «&. the 7rmeson field transforms under the representation R(1.O T '=(o °f) in a suitable representation.(^5Tip + xToTx) (35) (36) where %j> is the nucleon wave function with components ipp and U>N. and so (34) can be written in the form L" = g<j>. . and where \.In this case the interaction with the electromagnetic field is V = . (37) If invariance is postulated when the c^ are functions of position. Since the £.^ j ( * .c]kBakl) + Pea (39) .7 a * .e ( * o 7 a * 3 .Vjy*^) (38) where Bf: transforms under (37) as follows qBg = g(B?3 + clkBak] . and so is a selfdual antisymmetric tensor. Its interaction with nucleons and cascade particles is taken to be L" = igVa^jfaj This can be written in the form (33) L" = iglf<y5fty.*37«*o)A a (32) In this scheme.j of the vector representation in £ 4 .is the cascade particle wave function with components XB° a n d XHThus the above scheme gives the usual symmetrical interaction between the 7rmesons and the nucleons. that is T\ =  ( M 0 i + M23) etc. Under infinitesimal rotations in £4. satisfy the commutation relations for generators of the spin  representation in £ 3 .0) of the rotation group in £4. then it is necessary to add to £0 an interaction Lagrangian L i = . . we can take .$ (34) where T is the selfdual part of the infinitesimal generators M.
7 ( ^ M ^ . With the above identification it is seen that that the field equations (42) and (43) would then be exactly those for nucleons and cascade particles in interaction with the electromagnetic field4. Except that the mass difference between nucleons and cascade particles is not taken into account. and so T^. in such a way as to make all components of B°j vanish 3 except B„3. An interesting result of the above theory is obtained on comparing the interactions (32) and (38). It is seen that the electromagnetic interaction V can be considered as part of the 5field interaction L\ as long as q = e and Aa — BQ3.BMFM) =^ (43) where the fermion isotopic spin current density sfj is given by s?i = q{9i<ya9j*#"%) The total isotopic spin is T where (42) (44) f 3?3daa (45) (46) # = a 5 . as Shaw has confirmed to me. so that Editor's note: This is a mathematical error.BiwFg) The divergence of j g vanishes. This large mass difference seems to be the only real objection to the SalamPolkinghorne theory.q{B^Bpki . The total charge is then Q = T 03 . and apply such a transformation at each point in spacetime.The invariant free Lagrangian for the Ffield is L2 = FtfFiw where the 'field strengths' F£ are defined by Ftf = &*B% . then one could start from the Lagrangian L = L0 + L\ + L2.BfkB^) (41) (40) The field equations are easily derived and are (lada + m)Vi = qB?i'yayj and dpFtf + q(Biki0Fff . invariant under general transformations (38). In fact if the fifield were nonquantized.is conserved.d^B* . 4 3 .
Camb. Rev. Salam. 9 1 . Df = \{B^ .L. and so one would expect the existence of six i?particles of spin 1. 34. The interaction (38) can be written in terms of the nucleoli. N.. if the fifield does in fact exist. say. A. and so it is natural to suppose that the above six particles should exist in two sets of three with wave functions Ca and Da. [Nuovo Cimento 2 685.0) + i?(0.Ca The Cfield has the 7rmeson type interaction in the Salam scheme while the Dtield has an interaction of the Tmeson type (except that a vector interaction rather than a pseudoscalar one has to be taken). C and D fields. Rev.1). Proc.52a3) etc. it is not certain that it will be observed in its C and D forms separately. REFERENCES Kemmer. 191. Phys. C. J. 1938. 1955. However. Phil. k Polkinghorne. it transforms under the reducible representation i?(l. Yang. The interaction with the Cfield is q{iHaT^ and with the Z>filed is + XlaTX).] Schwinger. 713..while apparently starting with a more general theory.N. J.C. To be published. S o c . . where Cf = §(£& + B«3).354. cascade. k Mills. 1954. Of course the above reduction of Bjj to B03 could not be carried out in a quantized theory. for one would hope that the combination C3 + D3 would be the electromagnetic field. 1953. R. Since Bij is an antisymmetric tensor. Phys. 96. the usual theory is still obtained.
A similar argu. This duce a new field A? which transforms according to implies that one must consider transformations of the A? —> A. can be introduced as new field variie formations \p—>e \p. The theory may be reexpressed in terms of the Christoffel connection.. for example. therefore.. 2nd ed. then it is necessary to intro.„ appears in the matter field Lagrangian. p. We show that A 'JM were subsequently related to the Christoffel con. Weyl. and by a suitable (and uniquely determined) function of the earlier references cited there.the covariant and noncovariant conservation laws. it yields the familiar equations i?. These terms are almost certainly too small to make any experimentally detectable difference to the predictions of the usual metric theory. This Lagrangian is of first order in the derivatives. that the quantity r ' \ „ calculated from A "h was symmetric.> The extended transformations for which the 10 parameters become arbitrary functions of position may be interpreted as general coordinate transformations and rotations of the vierbein system. and to replace d^xp in the Lagrangian by coordinates as well as the field variables.. parison we give in Sec. 101. It is then unnecessary to introduce a priori curvilinear coordinates or a Riemannian metric. which include the metric to the corresponding group in which the ten parameters as well as the affine connection.to maintain invariance of the Lagrangian. although the covariant derivative of the metric vanishes. one actually requires the Lagrangian to be an invariant density rather than an * NATO Research Fellow. L.appear in the Lagrangian. involving variation of the coordinates as well as the field variables. to hit is changed.connection" A \. 96. B. Utiyama's discussion is extended by considering the 10parameter group of inhomogeneous Lorentz transformations. Hirzel. this means that. Yang and R.invariance) in terms of rectangular coordinates in flat formations specified by six parameters e*>. if the Lagrangian is invariant under phase trans.119 JOURNAL OF MATHEMATICAL PHYSICS V O L U M E 2. 191 (1954). i Thus. which will a "covariant derivative" (dM\ieA„)\p. and the simplest possible form is just the usual curvature England scalar density expressed in terms of lik* and A'. The free Lagrangian for the new fields is shown to be a function of two covariant quantities analogous to F„„ for the electromagnetic field. 1 See. and the new field variables introduced as a consequence of the argument include the vierbein components lit* as well as the "local affine connection" /1*. 4 we extend the discussion gravitational field variables. In Sec. 89. We shall follow the analogy with gauge transin order to do this it was necessary to introduce a priori formations as far as possible. and in that case additional terms quadratic in the "spin density" S*. Chap. but when matter is present there is a difference from the usual theory (first pointed out by Weyl) which arises from the fact that A '.in place of the restricted sixparameter group. I N T R O D U C T I O N I . Rev. but it also ment has been applied by Yang and Mills2 to isotopic means that only one system of coordinates is required. However. spin rotations. and for purposes of comcurvilinear coordinates and a set of 16 parameters hit11. 5 we consider the possible forms of the C.. N. the hk" were treated as given functions of x..„=0 for empty space. Leipzig. Mills. it is necessary nection r \ „ in the Riemannian space. Phys. Gruppentheorie una* Quanteninvariant. can be 1 vierbein components hk11. the affine connection r x „„ is nonsymmetric. The interpretation of the theory in terms of a of the gravitational field to the Lorentz invariance of Riemannian space may be made later if desired. 1960) An argument leading from the Lorentz invariance of the Lagrangian to the introduction of the gravitational field is presented... The new field variables become arbitrary functions of position. 1597 (1956). To make the action integral invariant. London.A P R I L . 2 new fields. 2 a brief discussion of linear Initially. Imperial College. 3 we discuss the invariance under Lorentz cussion to supply an argument for introducing the transformations. H. notably the electromagnetic field. and in Sec. space. The the Lagrangian. (Received August 19. T has long been realized that the existence of certain It is the purpose of this paper to show that the fields. transformations of the field variables. In the absence of matter. W. K I B B L E * Department of Mathematics. particularly with regard to system in a Riemannian space. factory procedure since it is the purpose of the disIn Sec. Rev. 2. Utiyama3 has proposed a method starting point of the discussion is the ordinary formuwhich leads to the introduction of 24 new field variables lation of Lorentz invariance (including translational A% by considering the homogeneous Lorentz trans. 3 Ryoyu Utiyama. free Lagrangian for the new fields. This is essentially but at a later stage they were regarded as field vari. 1931). multiply the invariant mechanik (S. As in the case of the 212 1. It is thus an attractive idea to relate the existence a priori. Phys. and is the analog for the vierbein formalism of Palatini's Lagrangian. as well as the "local affine related to invariance properties of the Lagrangian. and if we wish to make it in.— d„\. but this could to introduce 40 new variables so that a suitable covaonly be done uniquely by making the ad hoc assumption riant derivative may be constructed.necessitate some changes in the argument. so that the equation of motion relating A'. In particular. This is a rather unsatis.. though the emables and interpreted as the components of a vierbein phasis is rather different. and in that case yields a triplet of vector and that a Riemannian metric need not be introduced fields. 1961 Lorentz Invariance and the Gravitational Field T. and one must.a summary of Utiyama's argument. N U M B E R 2 M A R C H .ables analogous to A? if one considers the full 10paramvariant under the general gauge transformations for eter group of inhomogeneous Lorentz transformations which X is a function of x.
The change may and one m a y . One m a y define covariant currents by 7 ' M „ = _ (dL'/dA%)=(dL/dx. because the derivatives transform according to 5X.120 LORENTZ INVARIANCE AND THE GRAVITATIONAL FIELD 213 electromagnetic field. ( * ) } . which we regard as the elements of a column matrix x ( « ) .llTaX.y. 77. Rev. (2.^X^+A^TJC. define its covariant derivative be considered as a change of sign of e" and Ta. and the Ta are n given matrices satisfying commutation rules appropriate to the generators of a Lie group.„A\JILfi'0A''lA\. 7. and we shall assume t h a t this is so.„+e<'. The equations of motion imply n conservation laws X.. and multiplied by the gravitational constant. see footnote 3. though expressed in terms of an affine connection T \ „ which is n o t necessarily symmetric. Now. where Xr/J = 6>MX. New York). additional terms quadratic in &a. one can obtain a modified Lagrangian which is invariant by replacing Xr1 in I by a quantity X. fa. T h e y transform linearly according t o lTM=Ja'hTc. For a full discussion. I n fact. However. under t h e more general transformations of t h e form (2. 6 where F'^A%./dX i/1 )raX. with the Lagrangian L(*)=Z.fiu X .6). appear in the Lagrangian. This difference was first pointed out by Weyl. fa A. where the "currents" are defined by J^={bL/bX^TaX. =f) in Sec. T h e geometrical interpretation in terms of a Riemannian space is discussed in Sec.3 This is because with this choice of sign the analogous quantity for translations is 7>„ rather than — 7>„. and there is a cor. 4 a n d has more recently been discussed by Sciama. I n fact. when no m a t t e r is present it is symmetric as a consequence of the equations of motion.8) where L is regarded as a function of X and X.M m a y be called the covariant derivative of x with respect t o t h e transformations (2. LINEAR TRANSFORMATIONS We consider a set of field variables XA(x).{X(*). 7 (2. 4 5 .„.5) To do this it is necessary to introduce 4» new field variables A"? whose transformation properties involve £". Ui fiu~ fc Jb CL put 8 We have defined J'a with the opposite sign to that used by Utiyama. (2. 699 (1950). I n t h a t case.3) a (29) Unlike A „. therefore.6) then the condition (2. transforming according to D. T h e y are &4»„=tV.1) where the t" are n constant infinitesimal parameters. (2. The Lagrangian is invariant under these transformations if the n identities (dL/dX)TaX+ (aZ. This convention has the additional advantage that the "local affine connection" AijlL defined fa A.„)Tax. and this is done in Sec. From (2.c J'H = 0. Note that d/dx must be regarded as a row matrix. we choose t h e Lagrangian of lowest degree which satisfies the invariance requirements. 2. Phys.Vl'.5) determines the transformation properties of t h e new fields uniquely. I t satisfies the cyclic identity responding change of sign in (2. T h u s there is a difference between this theory and the usual metric theory of gravitation.2) and their covariant divergences vanish in virtue of the equations of motion and the identities (2.in an obvious manner. (2.tf do n o t cancel. I n fact. W. Sciama.2): 7'n = Tip _ J b f. Festschrift for Infeld (Pergamon Press. 5 I t arises from the fact that our free Lagrangian is of first order in the derivatives. We also consider linear transformations of the form SX=e'TaX. Weyl.„ which transforms according to &X. „ ^ = l1 fixjl a are satisfied. if one takes X.«%.„=0. with the hi* and A i'IL as independent variables. b u t otherwise it has an antisymmetric p a r t expressible in terms of the "spin density" <2". (2.ll=e°TaX. 4 specifies covariant derivatives according to the same rule 7 as r \ „ . 6. where we show that the free Lagrangian we have obtained is just the usual curvature scalar density..6) one finds X. .7) I n this way one obtains t h e invariant Lagrangian T h e expression X. (2. the Lagrangian is no longer invariant. * . Two covariant differentiations do not in general commute. I t is possible to reexpress the theory in terms of the Christoffel connection T\» or its local analog °A ' ' „ .4) and the terms in e°. to be published. the expression F%„ is a covariant quantity H.1). one finds SL=—^jJl'a.1). (2. b u t in which the parameters t" become arbitrary functions of position.^.„= taTaX:il.
)fa"cA%. and the index a is lowered with the metric 8 " g*b=fadfc b associated with the Lie group (except of course for a oneparameter group).„ are matrices satisfying From (3. 9 10 See L.(S. As before. KIBBLE where dL/dx" denotes the partial derivative with fixed X. Henri Poincare' 2. N o t e .I t is sometimes useful to consider also the variation at a fixed value of x.p(dL/dX)Sp„X+ (dL/dX)X.. the equations of motion m a y be used to obtain 10 conservation laws which follow from these identities. Thus the action integral over an arbitrary region is invariant if9 SL+L(5x") . as might be expected. Physik 5.7) (3. W. Ann. under a variation (3. T o obtain a strict conservation law one must sacrifice the covariance of the current. (3. the change in L is SL= (3L/dX)iX+ 8 (dL/dX. 1 which should be compared with (2..„. = 0. (3.2) I t remains to find a free Lagrangian L0 for the new fields. namely. Rosenfeld.3) one has 5X.„= (5X).1) (3. Clearly i 0 m u s t be separately invariant. since it is expressible in the form j>a= . Compare L."=x"+Sx1'. a n d it is easy to see 3 t h a t this implies t h a t it must contain Aa„ only through the covariant combination Fa„v. inst. (3. With the choice (2. 25 (1931).1) into 7'(fi)s f L'{x')\\d„x'»\\dtx. Then. 113 (1930). the equations of motion for the new fields are p f» * a . and the S.10) In particular. Ann. where ja=A\ftCaFc'"'.12) (2. = 0. B. but this can be absorbed by a trivial change of definition of vl°„ and 7V 811 The discussion here applies only to semisimple groups since otherwise gab is singular. —1). M =& 0 L+ (L8x») .5) where e and «""= — e"* are 10 real infinitesimal parameters.f{dL/dX^XM=0. are equivalent to the requirement t h a t L be explicitly independent of x.r>).6) Moreover. X(x)>X'(x') = X(x)+&X(x).3) where the tensor indices are raised with the flatspace metric ij" u with diagonal elements (1.4) for invariance of the action integral again reduces to 5 L = 0 . (2.(dLo/dA\.4) =J'f J a' Because of the antisymmetry of Fa1™ one can define another current which is conserved in the strict sense: ( / ' " . However.11) This is of course the typical transformation law of an invariant density.„) (5P„X. Note that (3.f)=0.121 214 T. it is obvious t h a t 5o commutes with 6V. X ^ . I t will be convenient to allow for the possibility t h a t the Lagrangian m a y depend on x explicitly. This extra current j " a m a y be regarded as the current of the new field A% itself. The action integral (3. AH t h a t is required is that it should be a scalar both in coordinate space and in the Liegroup space.8). and one could add to it terms of higher degree in F"^. since (bx* ) . — 1 .10). which express the conditions for translational invariance. (3.. ) . (3.MX. v d 1(0. and we shall assume t h a t they are satisfied.—nl»X. however. . p . t h a t it is not a covariant quantity. x"—*x.7).{dU/dA%)= . + J ' " . the condition (3.) These are evidently the analogs of the identities (2.IL)8XJl+ (dL/dx")dx\ There could of course be a constant factor multiplying (2. it seems reasonable to choose the Lagrangian of lowest degree which satisfies the invariance requirements.M = i £ ' » 5 p . I t is clear t h a t this Lagrangian is not unique.^ +Vl>fX. Bx"=e"^"+e". S<sX=X'(x)X(x) = dX&x"Xill. We now consider the specific case of Lorentz transformations. and yields the 10 identities 10 dL/dx"=L.„.e ^ X .10) of La. „ = 0 .8) (dL/dX.2). — 1 . 3. T h e simplest such Lagrangian is 8 Lo=lF%vFa"\ (2. LORENTZ TRANSFORMATIONS We now wish to consider infinitesimal variations of both the coordinates and the field variables. (I am indebted to the referee for this remark.)= J L(x)diX •'a over a spacetime region U is transformed under (3. Rosenfeld.„= £". M 1 6X = £e»»S„„X. whence SX.1).
p the action integral invariant.(dL/dX. v. to regard as inde. and for spinor components the Dirac matrices must be 7*. it includes contributions from terms in L We now turn to a consideration of the generalized which do not contain derivatives. momentum.6) This notation emphasizes the similarity of the t" t ransformations to the linear transformations discussed The condition (4.X. f ! . the currents in this case to be The reason for this is that we have not included the extra term L{5x"). (4.. the four functions £" specify field variables. 2. mentioned is of a different kind to those previously encountered.p«"pZ. However.„)Sp. (4. we make a of this fact. is not pendent functions e'"' and a covariant quantity. 6. The lefthand side of (3. which in this case is S0X=fd„X+it'"'(Sp.4) then follows from the identities in Sec. and we shall. We first eliminate the noninvariance arising from the fact that Xj. actually has the value p corresponding to the parameters t .„ term in (4. it is convenient to use Latin indices for e'1' further modification. It is of course not invariant under the generalized transformations (4. 12 Our A •'„ differs in sign from that of Utiyama.dp. Thus.MX.2) SX:fc= it''SijX. 5 V = . and angular momentum. k~ e'fcX. 7 the various tensor components of x must also have Latin indices. It is more suitable covariant derivative. / ' ^ • S V . which satisfies (and for the matrices £. and the additional term We now look for a modified Lagrangian which makes Sx'L. and Sp„\xpd„—x.3) On comparing this with (2.X. GENERALIZED LORENTZ TRANSFORMATIONS particular.4)./dX. It is instructive to examine these transformations in terms of the variation 50X also.„)X. and hence for the term h»pL in T^p.3) of the currents /M„.'1 Evidently.„)X. dX=ifi'Sipc (4.^i" (48) tities (3. The additional term just in the identities (3.p.8). in that it involves L and not dL/dXk. in terms of 50. (4. Thus it is clear that transformations (3.p+ieijSijX.3) by setting12 significance of the t" will be discussed in Sec.. by analogy with the fi7"_ _ tp Tit _ i . We function L it would be inconvenient to have two kinds can then impose the condition of indices.5) £" and x". one sees that the role of the matrices Ta is played by the differential operators and so the original Lagrangian transforms according to — 6\.' V ' .(1. and clearly equivalent. the differential operator X. t<".=X. I N V A RI A N C E AND THE GRAVITATIONAL FIELD 215 obtained by setting Xt = 5t''X. respectively.consider the problem in two stages. . &L+L{&x")./. Moreover. so that the coordinate and field action integral requires the Lagrangian to be an intransformations can be completely separated. (4.3 Compare footnote 6.£". For this reason. (4. a general coordinate transformation.8). the condition for invariance (3.4) 7« p = (dZ. In 4./dX.JXipr.T"phii^S"ii. In view variant density rather than an invariant. The original Lagrangian is then which determines the transformation properties of A"p " Note that since we are using Latin indices for S. replacing V by ?'.>).5) in which the parameters e" and we cannot remove it by replacing the derivative by a «"" become arbitrary functions of position. is responsible for the appearance of the term _. 5X. therefore. one could consider generalized transformations with Then. The geometrical and eliminate the 6''.4) is not simply 5oi=0. Thus the transformations under consideration are The first part of this program can be accomplished & K » = P . These transformations alone were considered (3.4) for invariance of the *'=0 but nonzero e'"'. We consider first the i'> transformations.4) since this avoids the explicit appearance of x. According to our convention. one might expect Note that it is J^p rather than r% which appears here..1).1). To do this it is necessary to introduce forty new by Utiyama.pSijX. The transformation of XiM is given by These are the conservation laws of energy. (•*•") r>„ must have a Greek index.p=^.7).„.lt+iei>. + ^ ' p .M+:MiVS'.k.„= keiisijX.dp)X. regard L as a given 11 function of X and Xk (no comma). retaining the Greek ones for 58'+£".7) and (3.„?'= 0. ' <Jn. and thus obtain an expression L' satisfying 5Z/0.p. but we shall later obtain an invariant expression by replacing Xk by a suitable quantity X.„ in (3. we shall convenient. definition (2.k or which transforms according to S0X=~^X.122 LORENTZ where T"p= (aZ. in the Lagrangian l where the A \= —A'\ are 24 new field variables. However.x.1) by replacing Xk in L by a "covariant derivative" X. because the condition (3. satisfying the iden5XM=i£i^.+xpd.
The analogs of the matrices Ta are clearly — a„ and Sij.„//*"e''*A/. it is defined for x and.6) we should add to X](i not a term in X b u t rather a term in X^ itself. I t is easy to see t h a t the only function of the new fields which obeys this transformation law. as in Sec.i&\L}. to (4. and satisfies the cyclic identity Because of the appearance of derivatives. we arrive at the same form for x. We shall call this the « covariant derivative.k transforming according. Originally.X.x.{(dL/dX.—iirkTk—§w2(£>2.8) of XM is already homogeneous.1. A i. and 1 in fact it is easy to see t h a t it is not. In other words.2).i. i ) (4. by a trivial extension for any other q u a n t i t y which is invariant under £" transformations. just like the second term of (2. As in Sec.17) «{x. 3 where the arbitrary constant factor has been chosen so t h a t § reduces to 1 when hk" is set equal to Sk". same sense as the modified Lagrangian L' of Sec.10)) b u t the corresponding modified Lagrangians differ b y = §fo"[(TM. _ L /  i A k . Thus. we can merely multiply by a new field: X.k=hk"X^. T h u s in order to define the modified Lagrangian ? completely it would be necessary to specify which of the possible equivalent forms of the original Lagrangian is to be chosen. b y simply ignoring the £*" transformation properties altogether.11) 1 I t should be noted t h a t the fields hk" and A \ are quite independent and unrelated at this stage. is itself an b\hi" = d/. This is equivalent to L2=irk. A k . = / ( • ' .15) (4.u T h e final form of our modified Lagrangian is To express the "conservation laws" which these currents satisfy in a simple form. Later we shall define another covariant derivative which takes account of f transformations also. b u t this is not true of the last term. rather different.„ri4 i 'tSiiXA'td^x. the first and last terms s can be combined in the form /ii*X.16) T h e n (4.w is inhomogeneous in the sense t h a t it contains X rather t h a n XM. 2.lA \t . (4. Consider for example the Lagrangian for a real scalar field written in its firstorder form Zi=7Tl>. since one must know how to treat the index on Xi>Here.t =5 t "x. of (4. it is convenient to extend the definition of the covariant derivative X^ (not Xk). ik i ii (4. . (4.„Aki„hklt. &»ij=2(d2/dA"ll)=&hk"(dL/dX:k)SijX. 13 Correspondingly. where hiJL = &k" — A"k. T h e y are SA 'll=t kA 'l.M+4 *•>*>] (4. billh/=iij.k<phTrkTrkW<p\ (4. 2.123 216 uniquely. We can easily obtain an invariant density ? ' by multiplying by a 1 suitable function of the fields already introduced: which is not an explicit divergence. and does n o t involve derivatives.18) The reason for this may be seen in terms of the variation where i 5oX given by (4. One can easily calculate the commutator of two i covariant differentiations.k)X.6) we should expect the covariant derivative to have the form This quantity is covariant under e'1' transformations.13) (4. therefore. Ife we R'W»IP+R . we simply ignore it. This means t h a t to obtain an expression X. Ji.k as that 15 obtained in the text. X . We have now found an invariant L'. though off course they will be related by equations of motion.^.yX. JVPI^+R'JWII»=0. 2. and transforms linearly under ciJ' transformations. B. is of course unimportant. 15 This gives Xi^Xi^iR^.12) t''kA l. one m a y define modified " c u r r e n t s " in terms of L=L{X.3) is to be expected. T h e reason for this 3 is t h a t if one starts with two Lagrangians L\ and Z 2 2 which differ by an explicit divergence.5) is satisfied provided t h a t § invariant density. W. is §=[det(/u*)]'. KIBBLE equivalent. for example.l.x. satisfying * (4.+ i i t T. the transformation law (4. ij (4. (We can drop the prime without risk of confusion. T h e reasons for this situation and the problem of choosing the correct form are discussed in the Appendix.„. where & „ is the inverse of hi".9) T h e position with regard to the last term in (4.}.„. then set A*1 *=hk'A"„. so that the presence of the derivative xtli in the last term a £>i.4).) I t m a y be asked whether this Lagrangian is unique in the . By analogy with (2.3) is. and are therefore 13 according to the t transformation law of hf. We wish to extend it to any quantity which transforms linearly under eij transformations. Note that this could not be done without extending the 14 r Multiplication of the entire Lagrangian by a constant factor definition./uvi%}=£>£{x. 5£+£V§=0. . we would have hi"u=hi?.k) by Zkll=d2/dhk"=^bi. T h e term involving e"'.14) Here the hk" are 16 new field variables with transformation properties determined by (4. then the modified Lagrangians Si and 8 2 are not necessarily equivalent.6) to be «/«*"= £"..
However. and should be this is that in evaluating X. ctii=(h.=0.. J C i *y)= ©". The reason for which is closely analogous to (2. It would also be possible to deduce from Eq. [^"(£VrtV>. that the invariant La must be a function only of the covariant quantities R'hi and C'H. P.•>= &4 ki„{hfhk°hk»hj'>)§A (unlike X((J) is such a quantity. Note that (5. (5.. Clearly 2o must be an invariant density.6) can be solved. See H. and footnote 5.15) would be There are in addition severa) quadratic invariants.8) is rather trivial. u one differentiates the h^ compared with (4. (5. 287 (1958)._ Q.10) in being only linear in the derivatives.3) therefore part of the "spin" of the gravitational field. (5. Compare footnote 13.=£**/£**<''• 5.lk=iR 'klS...5) (5. G. for example. (5..1) is not simply proportional to X.. (5. the ' 'conservation laws" can be expressed in the form6 {%Khk^+%kMu=\®"ijRiU @^i. for A *V In the simple case when © ^ vanishes. there are many possible forms for So.4) and this quantity does indeed satisfy (5.•. The most natural definition. 887 (1939).124 LORENTZ INVARIANCE AND THE G R A V I T A T I O N A L F I E L D 217 It is thus closely analogous to F"^. (5. (5. FREE GRAVITATIONAL LAGRANGIAN (4.t.16). Note that R'1^ is antisymmetric in both pairs of indices.. . As before. and if dinates do not appear explicitly. contract the upper indices with the lower. Thus one finds preted as the spin density of the matter field. Equation (5. Bergmann. In particular.). Physik 57. Eq. In terms of the e covariant derivative.19) (4. Rev. 261 (1929). See for instance V.12).a.uX. Z.5) a we set strict conservation law then it is easy to see. if we again choose for La the form of lowest possible degree.16 We now look for a free Lagrangian So for the new of angular momentum conservation in that the coorfields.v.„&#)=X'Jtf. Belinfante.7) differs from other statements involves X:i also. (5."hjvhj''hi'')bkli. in this case the expression within the parentheses itself vanishes. the condition that L0 be a scalar in two separate spaces is now reduced to the condition that it be a scalar in one space. In fact. As before. It is well known in the case of the ordinary metric theory of gravitation that many definitions of the energy pseudotensor are possible. Nevertheless.1) it appears to be a law of conservation of spin with no reference to the orbital angular momentum. so that (5. Fock. and therefore without i„{Ji»hk»hk»h?).However.2) corresponding "covariant conservation law" (4.7 may very reasonably be interof the index k... but there is a difference between this case and the previous one in that all the indices on these expressions are of the same type (unlike F°MU). this This quantity is expressible in the form quantity is not simply obtained by multiplying X„„—X„M by hk^hi". 112.6) one can immediately obtain a strict conservation law We now wish to examine the quantity X.7) is a rather surin X . 17 We choose units in which K— 1 (as well as c = ti— 1).8) but there is a considerable amount of freedom in choosing t*.„=0. rather (©*«+«". this means that there exists a linear invariant which has no analog in the previous case. 16 This is another example of the fact that for £» transformations derivatives play the role of the matrices T„.]. However.t . at least in principle. the orbital angular momentum appears in the Ri'ki=hk"hi"RiU (5. and one can. the covariant derivative of any quantity which transforms in a similar way to X may where be defined analogously. (5. and moreover adds an extra Aiklx term on account prising result. In fact. 19 18 which differs from (2. =§(At«C*. and i i C ki=(hk"hi"~hi"hk")b llll.6) From Eq. extending the definition of covariant derivative one can evaluate the commutator Xki—'X. Phvs.7) than X^. 20 The M ' ' „ are Ricci's coefficients of rotation. therefore.A^*«A. Now in particular X.i.. where however. the equations of motion for the new fields are $(&*. We shall not discuss the question of the correct choice of t*> further.jXC kiX. may be regarded as arising from this source.20). namely. by analogy with (4. Physica 6.19 It should be noted that Eq. then we are led to the free Lagrangian17 So=4§* (5. as this lies beyond the scope of the present paper. one finds20 Aij^ Aijll = iU ix\Ckij Cijk Cjki)y .8).20) With this choice of Lagrangian. J.k itself k «". but &ij.18 so that i i X. as one might expect. as in the case of linear transformations. See. since ©''.
x.10) is an explicit solution. also appears on the righthand side of this equation.3). It is important to notice that the derivatives of A"? in F"^ are ordinary derivatives.21 Thus the e'j transformations should be interpreted as vierbein rotations. X.3) with F%v given by (2.2) is consistent with the choice of metric (6.22 and from the tensors one can form corresponding world tensors by multiplying by hi/ or &V For example. Now the f transformations are general coordinate transformations. and may therefore be . and we must of course abandon for them the convention that all indices are to be raised or lowered with the flatspace metric ?. the Greek indices may be regarded as world tensor indices. not covariant ones.H—X. i§F*kiFatl. however. since one must preserve the invariance under the linear transformations.125 218 In general. The fields A"? should not be regarded merely as components of x when dealing with Lorentz transformations.=i>xi„irx„„D'i.10) T.„=A^Vl»=*^. J.u—Aij„vi. t = h ^ i X ^ + i A "uSijX+A^TtX). if we write then miShinkiSlu). is a symmetric covariant tensor. J. (6. Physica 7.) As before. GEOMETRICAL INTERPRETATION Up to this point.H) It may be noticed that the relation (4./i =X. We conclude this section with a discussion of the Lagrangian for the fields A"? introduced in Sec.lk then contains the extra term where F°uT"X. Note that v?= giitfl". B.1). This can be done provided that the matrices T" commute with the 5. Thus the quantity &»=6V*» (61) (5. Compare J. Otherwise. Vj\„=Vj. so that (5. We shall frequently use this convention of associating world tensors with given local tensors without explicit mention on each occasion. 2 when the "gravitational" fields hif and A''? are also introduced. from a local vector vi one can form v"=hi"vi. H. The commutator X. 239 (1931). this takes the form vi\„=v\v+Aijuvi. (6. is.„ may reasonably be called a "local aflfine connection" with respect to the vierbein system. for instance. It is easy to see that the scalar density § is equal to (—g)% where g=det(^„). 4. a condition which is always fulfilled in practice. and the simplest free Lagrangian for A". No confusion can be caused by using the same symbol v for the local and world vectors.1) and (4. because the affine connection is in general nonsymmetric. namely. one can see that any invariant function of A"? must be a function of Faki only. (5.„. while bk)i and A''? transform like covariant vectors. Math. and Phys. However. Weyl. and X.10) between X. *k <1=bi„vi\„= Vv—rv„!>\ where rx. It seems natural to define the covariant derivative of a world tensor in terms of the covariant derivative of the associated local tensor.1). 305 (1940).„ (6. we shall retain the use of two separate symbols because we wish to extend the definition of covariant derivative in a different way to that of Sec.2) one would form the world tensors corresponding to (6. since they are distinguished by the type of index. A.. s o that (6. To find the correct form of the Lagrangian.k is of the same type as (6. but it is useful to do so in order to be able to compare the theory with the more familiar metric theory of gravitation. then Skij is independent of A'1]. Now..11) hk" transforms like a contravariant vector under these transformations. 330 (1929). Schouten. 6.4) according to our convention. The field A '. Evidently.2). and the Latin indices as local tensor indices with respect to this vierbein system.9). therefore. v„=b\Vi. Then one finds that Xk in L should be replaced by a derivative which is covariant under both (2. one should consider simultaneously Lorentz transformations and these linear transformations. we have not given any geometrical significance to the transformations (4. Physik 56. Z. This gives fl\„=AiVl. since it specifies the covariant derivatives of local tensors or spinors.1).. or to the new fields Iik" and A1'). and indeed we have already used this convention in (5. W. K I B B L E interpreted as the metric tensor of a Riemannian space.2) and could be written simply as X . of a vierbein system in the Riemannian space.2) If the original Lagrangian L is of first order in the derivatives..1). respectively. to define the covariant derivatives of the world vectors (6. Belinfante.y.. Thus. hk» and J ^ are the contravariant and covariant components. in view of the relation (6. A'1]. x 21 22 23 (65) Note that this definition of T ^u is equivalent to the See for instance H. and use gn„ instead. The original field x m&y be decomposed into local tensors and spinors. 10.. and according to (4.23 For a local vector. It is moreover invariant under the tij transformations. (We shall see in the next section that the ordinary and covariant curls are not equal.
we shall assume in this section that L is only of first order in the derivatives.*)=£. 7. Then the tensor R^ is also nonsymmetric. „=a. One easily findsj first noted by Weyl.ii<. but not otherwise.11)) <g» —' pa.. are not just equal to derivatives of the "coordinates" hk". we can rewrite the covariant conservation is given by laws in terms of world tensors. Schrodinger. 8 only through the symmetric combination g"". and vice versa. The curvature scalar R has the usual formL R=R"n.„+§^ 'Aya+rV„S x "a > whereas the t covariant derivative defined in Sec. so that in general it is not the ChristoffelI connection. Sa^^'Sija+^Ji^a. The secondorder form of the Lagrangian may be obtained by substituting for Ai\ the expression (5. Thus there is. and1 M This is a generalization to nonsymmetric affinities of the result proved in the appendix to footnote 3. This gives I From Eqs. (6..12)I I t may be noticed that these are slightly more complicated than the expressions in terms of the 6 covariant derivative. the "momenta" A*1). and 'S is an additional term quadratic in Shtj.^r''x^r „+r'x„r\„. •/?«»= 2?Vm so that the free gravitational Lagrangian is just the usual one except for the nonsymmetry of r V . For a generic quantity a transforming according to ' J?„u=0. (This is the analog for world tensors of "A V ) Then R„„ is symmetric.11) as equations of motion.They are both world tensors. or in other words to M ' V Thus an interaction which appears simple in firstorder form will be more complicated if a secondorder Lagrangian is used. so that (5. investigate this difference in more detail.5) evidently does not guarantee that it is symmetric. where °S and °?0 are obtained from £ and So by replacing A% by °Aij„ (or equivalently T\„ by T V ) . fi^^fiv<* X 2**iu w pa. 1950). The situation is entirely analogous to that which obtains for any theory with "derivative" interaction. COMPARISON WITH METRIC THEORY For simplicity. 6Vi»=0(66) equations for empty space. See also footnotes 4 1 and 5.6) it follows that &«P=0. in which hk" and A ''„ are independent variables. In firstorder form. (6. (6.12) yields Einstein's familiar Ai\„=0.„. because hk" does not appear in obtained by simply omitting the last term of (6...u/1 = \R*'»uSija\RpwvZp"a— CV^x. I V is no longer symmetric.8).13). S'= 0 ?+°?o+ 1 ?. Thus one sees that i?%„ is just the Riemann tensor formed from the affine connection I V .„ is in a . we shall 3 that the commutator of two covariant differentiations.6) can be rewritten in the form Stf. (7. (6. ^=iS>(2SijkS""SiikSi'k+2SiihSi'k).126 LORENTZ INVARIANCE AND T H E GRAVITATIONAL FIELD 219 requirement that the covariant derivatives of the therefore equal to the Christoffel connection T V . the contraction where i?%„ and C V are defined in the usual way in terms of i ? V and C'u.p. and i (6..= K . since 3 there are only 24 components of A' '. namely.9)1 is then f.11).10). In the following section. i and can easily be expressed in terms of r V .13) one sees that in the absence: of matter the affine connection I V is symmetric. Thus Note that the two derivatives are equal for purely local[ the theory differs slightly from the usual one.5) and (5. and Eq..10) and (6.25 The equations of motion (5.. + C ^ . From (6. Spacelime Structure (Cambridge University Press. New York. general nonsymmetric.„r"„. : vierbein components should vanish.< . The difference between the theory presented here and the usual one arises because we are using a Lagrangian So of first order. only hk" and X are treated as inde . Finally. It is convenient to define a. and its antisymmetric part is given by the covariant derivative is defined by21 (6.It should[ be remarked that it would be incorrect to treat the 64 I components of I V as independent variables. no contradiction with the wellknown fact that the firstorder Palatini Lagrangian with nonsymmetric I V does not yield (6. the de5 finition (6. 25 See for instance E.IX _ ( . v ^ u . in the form24 since the covariant divergence of a vector density f" x i?%„=r%. (6. (6.„rv (6. 4 is.„=r\. However. In fact the T\„ are restricted by the 40 identities (6.*&.8) correspondingly the energy tensor density £.7) However.10) The conservation laws become •<.(35« = T _ T *~'ii^' pa •<^pa *~ap> so that it is consistent to interpret I V as an affine connection in the Riemannian space.14) c\.—a. (6.10) is an explicit solution for A''). when matter is present.1) In this Lagrangian. in a way 4 tensors or spinors.
This is the reason for the difference between this case and that of the linear transformations of Sec. B. therefore. We now wish to investigate the possibility of choosing a criterion which will select a particular form of L. This. Phys. W.26 which in the case of the scalar field discussed in Sec. B .S V O v would have a factor K_1.14) this is not equal to the ordinary divergence. ls so that the two cases differ with regard to the separation of the total angular momentum into orbital and spin terms. although it is true that the spin tensor obtained from Li is nonzero. P. 74 (195S). Polkinghorne.k — \rrir ({?. One can now see that in fact they differ by a covariant divergence. Here ?i is a "linearization" of Si' in the sense of T. and thus specify ? completely. on the other hand. extremely The conservation laws in the two cases are of course the small in comparison to other interaction terms. It was pointed out in Sec.1) with a negative be treated in this way. Now if we had not set K= 1. C. should not appearance of extra terms equal to (7. so that one would naturally terms. W.»»»V). a preferred position is assigned to the Hence we must conclude that for all practical purposes "wave function" <p rather than the "momenta" ivk. and invariant under all other transformations. ir . Kibble and J. differ in one important respect: Si is independent of Thus we see that the only difference between the two . because for some purposes <p and 7r'. In fact. so that it seems expect Skij to vanish. The most obvious criterion would be to require that the Lagrangian should be written in the symmetrized firstorder form suggested by Schwinger. 91. for a Dirac field. whereas the terms (7. Anderson.quantity Sha vanishes. Vv. This seems to be a resonable criterion. 27 26 . Nuovo cimento 8. then one would arrive at a footing. L. In this appendix we shall discuss the remaining ambiguity in the modified Lagrangian. Sciama for helpful discussions and if one writes the Lagrangian in its secondorder form in terms of <p only: comments. 713 (1953). which would select L\ rather than Li. whereas for L2 one finds interaction" terms.= (5 W .14) can be written in the form which yields the modified Lagrangian Si'=M? (#*>. The usual metric theory. Thus (4. but much smaller in magnitude. the two Lagrangians sign. equivalent to Si.•. However.».27 This should be contrasted with the secondorder form of 82. it is still true that the three spacespace components of the total spin but in view of (6. they would be proportional to tensor Ska has often been interpreted as the spin (see Appendix) density. It may be noticed that L\ is automatically selected Higgs.(§ft i "*>). Schwinger. but not for the vector k The author is indebted to Drs.1 (§A i *«') :i .1). There does not seem to be any really compelling reason for one choice rather than are zero. In par. It is clear that quite generally changing L by a divergence must change ? by the covariant divergence of a quantity which is a vector density under coordinate transformations.10). for L\ the theories is the presence or absence of these "direct.same. W. because the quantities Tki also differ.»*©»*V. KIBBLE pendent variables. then 8o £*. which is 82 . Now the ticular. If this Lagrangian were written in a firstorder form by introducing additional This corresponds to treating <p and irk on a symmetrical independent variables Ai'll. furnishes a possible impossible that they would lead to any observable criterion. and D. whereas ? 2 is not. 2. and clearly differs from 2Y by a covariant divergence. =4§. For. but there are plausible arguments for a particular choice. 4 that the generally covariant Lagrangians obtained from two equivalent Lagrangians L\ and Z2 are in general inequivalent. and the theory presented here is equivalent to the usual one.127 220 T.k<p.1 'JM. They are.1) would appear with the factor K. Correspondingly. because the matrices ACKNOWLEDGMENTS Sn are zero for the scalar field <p. J. In this way one achieves a vanishing spin tensor. but the arguments for it cannot be regarded as conclusive. is given by the Lagrangian Q"~ OWJOO another. With difference between the predictions of the two theories. The scalar field is normally regarded as Thus they are similar in form to the Fermi interaction a field of spinless particles. are eliminated from the latter by using (5. this choice. Rev. APPENDIX Li =\ip . 4 is without the extra terms (7. The equations of motion are equivalent to those previously obtained if the variables A*'.. Thus Li and L2 differ only in the values of the J. therefore. L=i(Li+L2). this may not in fact be the correct form identical to the one given here except for the choice. the derivatives are written on <p only. ¥>.
Fevnman and M.. Compare footnote 18. treating one of them line tp and the other like k. It yields Skij=aiJjka1fik.i: — iir:iyhf) m^Ap. however. but it should be mentioned because there are other grounds for treating \[/+ and i/_ on a nonsymmetrical footing.vanishes.128 LORENTZ 1NVARIANCE AND T H E GRAVITATIONAL FIELD 221 spin part of the (Oi) components of angular momentum. as Since the Lagrangian 8 must be Hermitian. and since F \ „ is nonsymmetric the covariant curl is not equal to the ordinary curl (though both 28 This form of Lagrangian may seem rather unnatural. 193 (1958). one could not write the derivative on \p alone. For a field of spin 1.only. The reason for the difference is that a.7. it might be questioned whether one should expect the spin terms to vanish even for a spinless particle. which yields the spin density Skij=ht.. 30 See R. the corresponding choice would be are of course tensors).kh$ ..J"M. In fact.kiyk{liy^mU +£§w2g*»a„a„.. Phys. 109. one easily sees that it is true in general that adding a divergence to L will change only the (Oi) components of 8. Rev. (A.l) with m = 0 would not be gauge invariant. the choice of Zi seems in this case to be the most reasonable.4.28 The modified Lagrangian may be expressed in terms of the world vector a.!.30 29 This has the rather strange consequence that for the electromagnetic field the "spin" tensor S^.29 For a spinor field \p. Indeed. \p±=h(lzhiyj)^. There remains.l) are covariant derivatives. since the Lagrangian is independent of . is here treated simply as a component of x. Even so. The difference is that the derivatives in (A. This gives the Lagrangian L=h^iyk(l+iy^ . GellMann. (A.and righthanded components. is introduced along with the gravitational variables to ensure gauge invariance.l) It should be noticed that the electromagnetic Lagrangian is not obtained simply by putting « = 0 in (A. Since it is not at all clear what significance should be attached to the separation of these components into "orbital" and "spin" terms. . P.:if4'iy'y4' which is again equivalent to the choice of the secondorder Lagrangian in terms of a. another possible choice: We could introduce a distinction between the left. symmetry between i£ and <p appears to demand that one should choose the symmetrized Lagrangian L = 5 (iiy V. which is a reasonable definition of the spin density. whereas .l).
a fact d i s cussed m a n y y e a r s ago by D i r a c .X^XjX^C. The b a s i c point i s the fact that electromagnetism is a nonintegrable phase factor. r e s u l t i n g in <PABCDA=I+fliV"Xk dx^dx"' where 1 J uv . It was pointed out by Weyl many y e a r s ago that the e l e c t r o m a g n e t i c field can be f o r m u l a t e d in t e r m s of an Abelian gauge t r a n s f o r m a t i o n .c o v a r i a n t quantity 4>a&y i s i m m e d i a t e 3 : ^adyK'=(K\R(U)\JHaByJ 6p • retain 445 (8) The function 6M*(x) defined on the manifold will be called a gauge potential. University of Wroctaw. N . Stony Brook. New York 11990 (Received 10 June 1974) A new integral formalism for gauge fields is described. (1) o r e m s a r e n a t u r a l l y developed.] D e fine a p a t h . . With t h i s definition additional c o n c e p t s and t h e  is not gauge c o v a r i a n t .—Consider a path ABCDA f o r m i n g the b o r d e r of an i n f i n i t e s i m a l p a r a l l e l o g r a m with s i d e s dx and dx'. (7) w h e r e R3ij is the adjoint r e p r e s e n t a t i o n for the e l e m e n t 4a. so that 3 <PA(A+**)=I + bll*(x)Xkdxi'. and m o r e r e c e n t l y by many a u t h o r s . o r gauge field s t r e n g t h . (4) /. It f u r t h e r allows a m a t h e m a t i c a l and p h y s i c a l d i s c u s s i o n of the g r a v i t a t i o n a l field as a gauge field. They a r e the F a r a d a y . f#vk i s . (PABCDA can be c o m puted by multiplying four p h a s e f a c t o r s like (1) t o g e t h e r . D e t a i l s will be published elsewhere. . to E i n s t e i n ' s .M a x w e l l fields whenG=£/(l). It is c l e a r that u n d e r (5) V A BCD A ~ V A BCDA . One might c a l l such f o r m u l a tions d i f f e r e n t i a l f o r m u l a t i o n s . The new f o r m a l i s m i s conceptually s u p e r i o r to the differential f o r m a l i s m and allows for n a t u r a l developments of additional c o n c e p t s . (2) ~X ' dX» •V*i J I'll (3) in which C. ») and c o n s i d e r a gauge G which i s a L i e group with g e n e r a t o r s Xk (k = 1. G e n e r a l i z a t i o n to other r e p r e s e n t a t i o n s R of G for a g a u g e . m). but not identical. w a s b a s e d on the r e p l a c e m e n t of '). and others. T h e a s s o c i a t i o n i s to have the group p r o p e r t y : cpABC = <PAB(PBC< w h e r e t h e p a t h s AB and BC a r e s e g m e n t s of ABC. Further developments are presented.—A gauge t r a n s f o r m a tion in the i n t e g r a l f o r m a l i s m is defined by a transformation <PAB<PAB = UVABZB' (5) w h e r e £A is an e l e m e n t of G which depends on the point A .—To .e. This idea w a s extended 1 in 1954 to the concept of gauge fields for n o n . but not identical with. .4x <PA BCD A 4A (6) Thus / )J „ ft ' = (fe!i? ad j!i>/„/. .„* will be called a gauge field. for G nonAbelian we have nonAbelian gauge fields. The s i m p l e t r a n s f o r m a t i o n p r o p e r t y (7) i s the definition for the concept t h a t / * is gauge covariant. [ F o r G = U(1) we have e l e c t r o m a g n e t i s m .—Consider a manifold with p o i n t s on it labeled by xM (jx = 1. <PAB will be called a gauge phase factor.A b e l i a n g r o u p s ./ is the s t r u c t u r e constant of G: X. including gravitation equations related to. . by 3„ . P e i e r l s . N U M B E R 7 PHYSICAL REVIEW LETTERS 12 AUGUST 1974 Integral F o r m a l i s m for G a u g e F i e l d s C. resulting in e q u a t i o n s r e l a t e d . . .. 2 . That formulation. Gauge transformation. . We s u m m a r i z e s o m e of t h e s e below.129 V O L U M E 33. Furthermore for an i n f i n i t i e s i m a l path A to A +dxu the p h a s e factor is c l o s e to the identity I oi G. Wroctaw. Gauge field strength./Xt. Gaugecovariant differentiation. Yang Institute of Theoretical Physics. Einstein's equations.. It i s the p u r p o s e of the p r e s e n t p a p e r to r e f o r m u l a t e t h e concept of gauge fields in an integral formalism. 2. and Institute of Theoretical Physics. like the Weyl f o r m u l a t i o n for e l e c t r o m a g n e t i s m .ieBn. State University of New York. nonintegrable) p h a s e factor <pAB a s an e l e m e n t of the group G a s s o c i ated with path AB between two points A and B on the manifold. Poland.d e p e n d e n t (i. 2 This fact i s now g e n e r a l i z e d a s follows: Definition of a gauge field.
(15) and (16) of Ref. = K. /. ]i II X = 0 (conserved c u r r e n t ) . a s o u r c e f o u r . Now w e i n t r o d u c e a m e t r i c for it and d i s c u s s a r b i t r a r y c o ordinate t r a n s f o r m a t i o n s .1 ' X II rJXlinv = r w h e r e the s e m i c o l o n r e p r e s e n t s (12) Ra (17) 0 which i s s a t i s f i e d by all gauge fields on all R i e mannian manifolds. (a) C o n s i d e r the n a t u r a l R i e m a n n i a n geometry of a s e m i s i m p l e L i e group. I t s p a r a l l e l ... A n e c e s s a r y and sufficient condition for a p u r e s p a c e is R„ .d i s p l a c e m e n t gauge field i s s o u r c e l e s s and a n a l y t i c . Thus we have the following: Theorem. leading to the concept of " t o t a l c h a r g e . The index A h a s ? n2 v a l u e s and we w r i t e k = (ajl)..X • (18) f J]iv k II X . It i s e a s i l y shown that /ui. This gauge p h a s e f a c t o r which i s itself an element of G gives a gauge field which i s analytic and sourceless. NUMBER 7 PHYSICAL REVIEW LETTERS 12 AUGUST 1974 gauge covariance in differentiation we define (9) t i a l and gauge fields a r e r e s p e c t i v e l y (KB t\=d£+b*(K\Zb\J)r.0 .—For any R i e m a n n i a n manifold. A n i n t e r esting t h e o r e m i s that fiiv\\ +fv\\ii +A(ii< . Source of gauge fields. in analogy with e l e c t r o m a g n e t i s m . E.—So far we need no m e t r i c f o r the manifold. 1 this w a s E q . " Parallel displacement gauge field.—A R i e m a n n i a n manifold for which the p a r a l l e l .t and VB.—Parallel d i s p l a c e m e n t defines a gauge field with G being GL(w). A fourdimensional Einstein space one for . One c a n a l s o g e n e r a l i z e E q s .d i s p l a c e m e n t gauge field is s o u r c e l e s s will be called a p u r e s p a c e . (14). Define ipAB a s that for an i n f i n i t e s i m a l path A B . (16) T h u s by definitions (9) and (11) '  > II M _ V + )fia(' (11) (otBl dtji' k f i>\ ^ a IIM + 0 P ° f \rl(7E) * 11V JUV II X J]iv\> MX f J CCV e t c . W e have «M It i s i m p o r t a n t to r e c o g n i z e that in t h i s definition w e have c h o s e n a fixed c o o r d i n a t e s y s t e m . In t h i s connection we o b s e r v e that f o r GL{n). MAB i s a r e p r e s e n t a t i o n of an e l e m e n t of GL(w).v e c t o r J^ for a gauge field: J„ x the u s u a l R i e mannian c o v a r i a n t differentiation with a and /3 t r e a t e d a s u s u a l c o n t r a v a r i a n t and c o v a r i a n t i n dices. T h u s p a r a l el d i s p l a c e m e n t i s defined by a n B X « m a t r i x MAB which gives t h i s l i n e a r r e l a t i o n s h i p . a linear r e l a t i o n s h i p between any v e c t o r VA at A and i t s p a r a l l e l v e c t o r VB a t B.— We define. W e c o m e then n a t u r a l ly to Riemannian covariant quantities and doubly covariant derivatives.g. In o t h e r w o r d s MAB — NAMABNB'1. Comp a r i s o n with (5) shows thus that a c o o r d i n a t e t r a n s f o r m a t i o n g e n e r a t e s a s i m u l t a n e o u s gauge t r a n s f o r m a t i o n of t h e p a r a l l e l .„* i s doubly c o v a r i a n t . So f a r only nonanalytic e x a m p l e s a r e known. (10) which i s the g a u g e . t h e u s u a l n o n l i n e a r t e r m in the t r a n s f o r m a t i o n of {£y} i s p r e c i s e l y the nonlinear t e r m needed in the gauge t r a n s f o r m a t i o n of the gauge n o n c o v a r i a n t quantity & M (aB) .'lIX +/l'Xllrifi +A11II. J IIV\\\0J1'.—Gauge fields for which f]iVk*0 and J* = 0 a r e of p h y s i c a l i n t e r e s t .i (19^ i. The rule works also in general. Gene r a l i z a t i o n t o o t h e r c a s e s i s obvious.f Jfiv (13) After s o m e computation one d e r i v e s a t h e o r e m : g'^J. (b) C o n s i d e r t h e s a m e R i e m a n n i a n manifold Oi a g r o u p G a s above in (a). (14) which in e l e c t r o m a g n e t i s m s t a t e s c h a r g e c o n s e r vation. t h e i m p o r t a n t concept of p a r a l l e l d i s p l a c e m e n t defines.d i s p l a c e m e n t gaucc potential. ')0Mp J vv ~ Bpi' • (15! w h e r e Zk i s t h e m a t r i x r e p r e s e n t a t i o n of Xk. 1. Pure spaces. along any path AB. s i n c e <pAB i s c o o r d i n a t e .s y s t e m independent.e.130 VOLUME 33. In Ref. Introduction of a Riemannian metric. bvk i s R i e m a n n i a n c o v a r i ant. In fact. 4 We now c a n c o n s t r u c t two g e n e r a l t y p e s of gene r a l types of e x a m p l e s . <pAB= (A _ 1 B) 1 / 2 .B i a n c h i identity. The gauge p o t e n 446 Nontrivial sourceless gauge fields. A c o o r d i n a t e t r a n s f o r m a t i o n would g e n e r a t e a l i n e a r t r a n s f o r m a t i o n in the v e c t o r s p a c e s V.
Choose Ckab to be the structure constants for GL(«).—The electromagnetic field and the usual gauge fields are special cases of gauge fields. Utiyama. Acad. 349. The resultant equations a r e satisfied by (15) and (19).and R. edited by H./_. p. 96. Wu and C. is a pure space. Gravitational field as a gauge field. where » / L(b Iftxf . 3 We use the summation convention for repeated indices. Bull. (22) in which b ( a S ) and g a r e independently varied. BiafynickiBirula.? d"x=0. Ser. Math. 1969). in Properties of Matter under Unusual Conditions.131 PHYSICAL REVIEW and g LETTERS Now form the variation (ctB) . A pure gravitational field is then described by a pure space as defined above. had concluded that Einstein's equations a r e gaugefield equations. Sci. Pliys. (21) which of course also contains derivatives of b. JJ.v = 0. 5 R. given by (16). 25 (1962). N. It is a great pleasure to acknowledge the warm hospitality extended to me during my visit to the Institute of Theoretical Physics at Wroclaw where this paper was written. Yang. (New York) _19 1. satisfying (12) and (13). Rev. where M is the dimension of a representation of G. We believe that was an unnatural interpretation of gauge fields. _101.* is given by (3). 'C. New York. Fernbach (Wiley. N. Upper case Latin indices run from 1 to M. There a r e advantages in this identification and we shall come back to this topic in a later communication.k 447 . NUMBER 7 12 AUGUST 1974 which RaB = 0. Yang. 1957 (1956). L. Write the L of (20) as a functional of b ^ andA' x ": L = Hb„(aR\gXv). A natural question is whether one should identify these same equations for the paralleldisplacement gauge field as the equations for the gravitational field.. Phys. I. Mills.. Greek indices run from 1 to n. Mark and S. Astron.. Pol.—Equation (13) with JM* = 0 follows from a variational principle 5(/£•(/". m of course is also the dimension of the adjoint representation of G. If one adopts this identification then gravitational equations a r e thirdorder differential equations 5 for g#v. Phys. 4 T. Rev. Sci. 135 (1963). One could also find a variational principle which is satisfied by a pure space (19). Lowercase Latin indices run from 1 to m. Phys. L=fJfaB'gmg"*Cj'C1 (20) In the variation gvv is kept fixed and 6 p is v a r ied. Ann. 2 S. Ckah a r e not varied.gXv)L VOLU ME 33. T. and/M[. 191 (1954). Variational principles. Mandelstam.
and reveal a very bold departure. It has been pointed out. that there is a significant difference between the. generalized forms of the Ward identity are obtained between certain vertex parts and the selfenergy. 745 (1959). however. Bogoliubov. In the BCS model.132 PHYSICAL REVIEW VOLUME 117. NUMBER 3 F E B R U A R Y 1. and the nature and effects of such collective states are discussed. D. 34. Nuovo cimento 7. Buckingam. 1958). as was first derived by Bogoliubov. Phys. 496 (1958). Schafroth. invariant character of the theory. Pines and R. Wentzel.3 In the Meissner effect one deals with longitudinal current. Rev. 91 (1959). Rev. and 0) refers to the superconductive ground state. Phys. 1. from the conventional approach to manyfermion problems. 407 (1958). 781 (1958).2 which can be excited only by the longitudinal current. Such "quasiparticles" are not eigenstates of charge and particle number. ibid. can be maintained in the quasiparticle picture bv taking into account a certain class of corrections to the chargecurrent operator due to the phonon and Coulomb interaction. In fact. Bardeen. Valatin. S.S. has also reformulated his theory as a HartreeFock approximation. Theoret. 73 (1958) [translation: Soviet Phys. Chicago. Anderson. The Meissner effect calculation is thus rendered strictly gauge invariant. the second term vanishes in the limit q — 0. Letters 2. Nuovo cimento 10. Atomic Energy Commission. 51 (1958)]. Matsubara. 1900 (1958). ' ' 2 latter. Rev. 41. as is obvious from the viewpoint of the quasiparticle picture. Rev. 3 M. Rickayzen. Progr. approximation. 1763 (1957). 827 (1958). Bogoliubov.4 We will develop this point a little See also J. Blatt and T. 236 (1959)].2 who introduced coherent mixtures of particles and holes to describe a superconductor. Schrieffer. Moscow. later as the main part of the paper. The author is indebted to Prof. Namely.transversal and longitudinal current operators in their matrix elements. Letters 2. N. that the former is a necessary consequence of the 108. J. presents us with a remarkably good understanding of the general features of superconducivity. inherent in the BCS theory. and Shirkov. G. 1765 (1957). U. 33 (1959). Phys. Phys. The University of Chicago. but essentially keeping the BCS result unaltered for transverse fields. Rev. 549 (1959) [translation: Soviet Phys. Rev. 731. I l l . 72 (1958). 34. Phys. Nauk 67. Nuovo cimento 5. Rev. the original result will remain essentially correct. external vector potential A and the induced current / . Uspekhi Fiz. W. G. If such collective states are essential to the gauge* This work was supported by the U. Blatt.R. 110. K. Phys. A mathematical formulation based on the BCS theory has been developed in a very elegant way by Bogoliubov. Phys. N. I960 QuasiParticles and Gauge Invariance in the Theory of Superconductivity* YOICHIRO N A M B U The Enrico Fermi Institute for Nuclear Studies and the Department of Physics. 58. creates at the same time certain theoretical difficulties which are matters of principle. Yosida. however. Letters 2.S. and poses a serious problem as to the correctness of the results obtained in such a theory. Progr. and May. Illinois (Received July 23. En (1. Bogoliubov for sending him a preprint. to the first order in the external electromagnetic field. The form of the equation implies the existence of a particular solution which does not follow from perturbation theory. A HKii{q)Aj{q). A New Method in the Theory of Superproblems is to recognize it as a generalized HartreeFock conductivity (Academy of Sciences of U. On the whole. and Schrieffer.S.1) (0\ji(q)\n)(n\ji(q)\0^ + En 648 . the BCS theory. which leads to the existence of an energy gap. I l l . Cooper. This. N . 843 (1958). R. Phys. Thus the derivation of the Meissner effect in the original BCS theory is not gaugeinvariant. Phys. but cancels the first term (the a linear relation between the Fourier components of the longitudinal sum rule) to produce no physical effect. NUMBER of papers have appeared on various aspects of the BardeenCooperSchrieffer1 theory of superconductivity. further here since it is the starting point of what follows Nuovo cimento 5. Theoret. then one might argue 1 Bardeen. (Kyoto) 21. 162 (1957). 1488 (1958). Bogoliubov.S. the second term does not vanish for a This question of gauge invariance has been taken up by many people. But this point has not been clear so far. J. J. leaving > the first term alone to give a nongauge invariant result.R. P . I l l .—Uspekhi 67. Another way to understand the BCS theory and its Toimachev. 1 Recently N . The gauge invariance.. Exptl. Matsubara. whereas for a transversal field. one can write down an integral equation defining the selfenergy of an electron in an electron gas with phonon and Coulomb interaction. Rev. As a result. Theoret. (Kvoto) 20. there exist collective excited states of quasiparticle pairs.(ff)  «><»  iy( —ff)10> *«(?) = —<O]PO)S n \ p and j are the chargecurrent density. 1175 (1957). In an approximation which corresponds to a generalization of the HartreeFock fields. and discussed the gauge invariance collective excitations from this viewpoint. J. It is shown also that the integral equation for vertex parts allows homogeneous solutions which describe collective excitations of quasiparticle pairs. Phvs. 112. and which leads to the energy gap equation and the quasiparticle picture analogous to Bogoliubov's. M. 817 (1958). M. 1959) Ideas and techniques known in quantum electrodynamics have been applied to the BardeenCooperSchrieffer theory of superconductivity. G. Phys. 106. INTRODUCTION which is given by the expression Ji{q) = with <01 i.
The justification of such a procedure may be given by writing the original Hamiltonian as namely.5) 4>{xy) » e aw+ft(»V(ry)." and the diagonal part of Hs represents the selfenergy (or the Hartree potential) for such particles arising from its interaction with the vacuum. The significance of the BCS theory lies in the recognition that with an essentially attractive interaction V. He derived this requirement from the observation that /7i n t' contains matrix elements which spontaneously create virtual pairs of particles with opposite momenta. we often take a picture in which particles and holes can be created and annihilated.2) K is the kinetic energy plus any external field. (1.g.y)fr(y)fi(x)<PxcPy = i7o+#int.3) + =H0+HS. and cause the breakdown of the perturbation theory as the energy denominators can become arbitrarily small. (1.3) will change x and 0 according to x(xy) * eiXM+A(y)x(xy).6) J J i.133 QU A S I . The diagonal part of Hs is still arbitrary.4). The distinction between Eqs. Goldstone. as applied to such pair creation processes. In this way we can interpret Ho' as describing single particles (or excitations) moving in the "vacuum.53(. and the corresponding ground state has a lower energy than the normal state. For the analysis of the Hartree approximation in terms of diagrams. 5 Equation (1. as in relativistic field theory. We can fix it by requiring that <l'ff t o . (1. (1.k +ti (x)<f>ik (xy)tyk (y) +\f>k {x)<f>ki+ {xy)ipi (y) ~]d3xd3y (1.'l) = 0. This condition is contained in Eq. determines only the nondiagonal part (in quasiparticle energy) of Hs in the representation in which Ho{Hs is diagonal.•(*)—» ealr)4'i(x) applied on Eq. The condition (1. We diagonalize H0' and take. x and <p will in general depend on the external field as well as the interaction between particles. Proc.v.6) and demanding that Hint' shall have no matrix elements which would cause singleparticle transitions. i= 1. \V{xy)^k+{y)h+{x)).5) was first invoked by Bogoliubov2 in order to determine the transformation from the ordinary electron to the quasiparticle representation. see J.„ t '0)=<nff0)=0. spin up and down along the z axis). (1.P A R T I C L E S IN S U P E R C O N D U C T I V I T Y 649 Take the Hamiltonian in the second quantization form for electrons interacting through a potential V: J . .5).5) and (1.k Xrh+(y)V(x. the condition (1. <f>+(xy) —» e~AM~iX(v)<t>+(xy).e.4) V(xy){+k+(y)Mx)}. (1. It is also separated from the excited states by an energy gap ~ 2 $ . We write thus a linearized Hamiltonian Ho'= f E 4>i+K«fr4Px+ f f E bPi+(x)xik(xy)My) J i + if in  n) more than one particle change their states from those in 0). On the one hand. The HartreeFock method is equivalent to linearizing the interaction Z7int by replacing bilinear products like ipi^ix^kiy) with their expectation values with respect to an approximate wave function which. is determined by the linearized Hamiltonian. ^k(xy) = •i>ik {xy) = + iV(xy)(My)Mx)}.=i J J i. Equation (1. 2 refers to the two spin states (e. These two together then are equivalent to Eq.6) is not so clear when applied to normal states. The defining equations (1.6 Since in manybody problems.6). This is because a phase transformation \p.. Transitions between occupied states or unoccupied states are given by Eq. in turn. for example. there is little difference when one particle is added just above the ground state..4). (1. where XafoO = a. on the other hand. the groundstate eigenfunction which will be a SlaterFock product of individual particle eigenfunctions. (1.4) then represent just generalized forms of HartreeFock equations to be solved for the selfconsistent fields x and <f>. We may consider also expectation values (ti(x)>pk(y)) a n d (<Pi+(x)^k+(y)) although they would certainly be zero if the trial wave function were to represent an eigenstate of the number of particles. particles and holes (negative energy particles) are not separated by an energy gap. i.5) refers only to the transitions from occupied states to unoccupied states. the vanishing of the diagonal part of Hint for the states where one more particle (or hole) having a Hamiltonian H0' is added to the ground state. a nonvanishing <j> is indeed a possible solution.5) should also be interpreted to include the case where \n) and 0) differ only by such pairs. There is a complication due to the fact that they are gauge dependent. we demand our approximate eigenstates to be such that <nff. as is the case for the true wave function. no matrix elements which would effectively modify the starting H0': to put it more precisely. In the above formulation of the generalized Hartree fields. (1.ry) fv(xz) E^/Hz)^))^ (1.
any calculation based on a particular < is open > / We start from the Lagrangian for the electronto question.\j+< Xp)Mp)+i (phMp)l invariance would be to find the dependence of <j> on the +E* Km*(k)*<p(k)<p(k)l external field explicitly. Guided by the wellknown theorems about gauge invariance. Ward. 182 (1950). Then by perturbation. the Hartree approximation. the The Lagrangian then becomes: gauge invariance is thus strictly established insofar as effects linear in the external field are concerned. are obtained. 6 J. 267 (1957). we have included in h(k) the frequency factor: h(k) = hi(k)h. 6 They are intimately related to inherent invariance properties of the theory. Later we will make remarks whenever necespart which corresponds to the Hartree approximation. leading to analogs of the Ward identity. 7 the bound pair states are drastically modified. A natural way to reconcile + + the existence of 4>. with the momentum k (energy primary electromagnetic interaction when <f> is first a>k=ck) running up to a cutoff value km(um). each other. 78. Rev. and the existence of an energy gap is immediIt will turn out to be convenient to introduce a twoately recognized. Kinoshita and Y. Anderson [Phys. 94. •s/V r. ind cluding the Meissner effect. solution. We will first develop the The Coulomb interaction between the electrons is FeynmanDyson formulation adapted to our problem. Rev. not so in other gauges. A very \ at / interesting result emerges when we observe that one + E *L>(*)*(A)eV(A)*>(*)] of the exact solutions to the vertex part equations h 1 becomes a homogeneous solution if the external energy~g E ^+(p+k)r3^(p)h(k)^(k)+i: ep momentum is zero. and their dispersion law determined. 3 P.k (1.1) longitudinal potential A — — gradX. which is supposed to be uniform and 7 effect because it can be transformed away before making isotropic. Roy. W. If the gauge invariance can be l maintained. which are seemingly contradictory to uncomfortable. FEYNMANDYSON FORMULATION gauge.2) are led to consider the socalled vertex parts. C. Soc. turning We use the units h= 1. which we want to keep. This should not be done in an arbitrary manner. Thus the existence of the bound state is a logical consequence The last infinite cnumber term comes from the rearof the existence of the special selfenergy <j> and the rangement of the kinetic energy term. but by studying the actual influence of Hini on the p is the phonon field. even into the plasma modes due to the same mechanism as if <j>(xy) = 53(x—y) times a constant in some gauge. and expresses a bound state of a pair with zero energymomentum. Phys.8 (1) is the volume of the system. not included for the moment in order to avoid comand write down an integral eauation for the selfenergy plication. Therefore. (2. 598 (1953). Nambu. 1900 (1958)]. certain exact and the corresponding 2X2 Pauli matrices solutions are obtained in terms of the assumed selfenergy part. This is certainly gauge invariance. e„r 3 I * ( # ) £ = E*+(i>) I i Later we look into the collective excitations. This situation will also be studied. it is in the normal case.6). gh{k) represents the strength points raised here by means of the techniques developed of coupling. unless we can show explicitly that physical quantities do not depend on the 2. When an integral equation for the general vertex part is written down.134 650 Y OICH I RO NAMBU It is especially serious for <j> (and <j>+) since. (London) A239. other bound states with nonzero energymomentum = £o+£mt+const. ep is the electron kinetic energy relaAfter these preliminaries. sary about the modifications when the Coulomb interIt is observed that it can possess a nonperturbational action is taken into account. c is the determined without the external field. Among other things.k P 8 For convenience. we are going to study the tive to the Fermi energy. component notation 9 for the electrons Next we will introduce external fields.) in quantum electrodynamics. has also introduced this twocomponent wave function. When the Coulomb interaction is taken into account. Rev. . the dependence must be such that for a g E >Pi+(p+k)Mp)Kk)<p(k). with gauge £ = E * E. Compare also T. which include the "radiative corrections" to the primary chargecurrent operator. The fields obey the standard commutation relations. \/V p. we *(*) = ( ) or *{p) = [ ) . it reduces to Eq. but will not be important except for the calculation of the total energy. (2. I t would not be enough to say that a longitudinal electromagnetic potential produces no phonon system. 112.. phonon velocity. Phys.
In the manyparticle system.¥. unlike the case of quantum electrodynamics. With the aid of these Green's functions. Here we neglect the renormalization term Si since the two conditions (2. we will ignore the small imaginary Having defined the vacuum. Schwinger. Rev. P. Namely the second order selfenergies coming from the phononelectron interaction have to be cancelled by the first order effect of These second order selfenergies are represented by the nonlocal operators12 (Fig.tf) £=(£o+£s)+(£int£. Rev.*j+(p')} = dij8I. Phys. (2. A0(xl) = Ga(ppo)=i\P f <*.10 The unper. Aoikkde*^^'(Pkdh. we are able to calculate the 5 matrix and other quantities according to a welldefined set of rules in perturbation theory. In the same spirit S should actually be in the form 2i(/>)^or2. We get for their Fourier representation (in the limit V — oo)IOa > ^=T. we have {**(*). (2. lumping together all turbed ground state (vacuum) is then the state where topologically equivalent processes. R. Acad. using represent electron and phonons. MoQ<L. We now determine 2 and n selfconsistently to the second order in the coupling g. 75. there will be no ultraviolet divergences.k0 = ait) = 0. 769 (1949).k. Feynman. respectively.1736 (1949). these energies express (apart from the selfinteraction of the electron) the average interaction of a single particle or phonon placed in the medium. 455 (1951). 76. f~~\ FIG.+(*V))>= [Go(xx'. We are. Let us thus introduce the approximate selfenergy Green's functions for free electrons and phonons Lagrangian £. tOla.7) The free electrons with "spin" functions u and phonons obey the dispersion law Lo(p.135 QUASIPARTICLES IN SUPERCONDUCTIVITY 651 Especially for ~$.10) (2TT)4 J\k\ pO—tpTz = i(po+*pT3)/(p<ii—ep>+ie).p0=Ep)up=0. Schwinger's: Proc.9) (2. These selfenergies may be obtained in a perturbation expansion with respect to Hint. But to the extent that the singleparticle picture is the occupation number. Natl.) are easily determined. 452. and write <r(* 4 (*0. 1. 74.(^).8) Go(xt) = (l/(2ir) 4 ) f G0(#>o)e''» •*{'»'d>pdpo. {*i(p). 1) S((*+0/2)= f f f*+(xt)S(xx'.po=ep)up=0. 10a P stands for the principal value. M(k. It is true that the selfenergies are in general complex due to the instability of single par10 F. J. Sci. iir s." II will consist of two parts: n(£o£) = ni(£)£o 2 +n 2 (£) in conformity with the second order character (in time) of the phonon wave equation. it in the denominator is a small positive imaginary quantity. Aa{kh )=i\P i: iir8(ka2—c2k2) I k02C2k2 ] = i/(k02cW+u).v^P+L^v+T. makes physical sense. 486.<S>{x't')d?x<Px'd{tl').to be interpreted in the sense of Feynman. the timeordered part of the selfenergies in the following considerations. Phys. Dyson.5) 3.k0=nk) = 0.. Although we followed here the perturbation theory of Dyson.11) <P((/+0/2) = i f J [<p(xt)P(xx'. all individual electron states e p < 0 ( > 0 ) are occupied (unoccupied) in the representation where >P<ri~(P)4'i(p) tides. S. Rev. interested in the Hartree method which proposes to take account of them in an approximate but nonperturbational way. Phys. themselves being under the influence of the selfenergies 2 and II. tt') X<p(x't')dsxdsx'd(lt'). Second order selfenergy diagrams. there is no doubt that the relations obtained in this paper can be derived by a nonperturbational formulation such as J.gn(T3ep)S(po— T3ep) (2.khfkM^. M0n=M. J. L(p.13) can be met without it.v(x't')))=A0(xx'. 1439 (1948). Because the phonon spectrum is limited. 37.ll') Y. A{kh) = i/(M(kk0) + ie). u We use the word nonlocal here for nonlocality in time. We will analyze in particular the selfenergies of the electron and the phonon. 11 . (2. U. however.p.* J r + (y)}s*<(*)* i + (y)+*/ . (2. Solid and curly lines We may now formally treat Hint as perturbation. L0X=L. (2.90 whereas they obey in the medium 2 will be a function of momentum p and "spin.11 The propagators corresponding to these modified electrons and phonons are G(pp0) = i/(L(pp0)+i sgn (/>„)«). (y)* < (*) = M'(*y). (26) (T(v(xt). All diagrams are the formulation of Feynman and Dyson.
P(kko)=ig h(khyf 2 + tf^ + J£TA + /fi°fa +.Its eigenvalues are tions may be written (2.''™''' for %+ and ^ in (2.13) defines an infinite sum of a particular class of diagrams. will turn out separated by a gap ~2\<j>\. so that no offdiagonal part exists.6) mean in the present case that 2. Thus.16) theory.13) only special matrix elements of both sides. we have to adopt the hole picture and conclude that the n(kn*)=.'. and offdiagonal conditions will fix these. 1 Up'. (2. (2. The pair creation of electrons is possible because Sfr. 2(pO)ND=<S (pO)ND( (2. for a fixed momentum. n(ko)=p(ko). p.13) Since Ho' describes by definition excited states. If 4> remains finite on agreed to omit possible imaginary parts in 5 and P. Since II consists of two parts.)up„ while the nondiagonal part corresponds (2. po+koWpdpo. (2. In the latter..11 It is important. (2. With this understanding. A=lEp+Ho'(p)y2Ep. and (b): for the nondiagonal matrix elements for creating a pair out of the vacuum.. the positive and negative states are (The nondiagonal components.0)up.11) we have chosen more or less arbitrarily {t+t')/2 as the fixed time to which we refer the nonlocal operators S and (P.P(kQ*). ground state (vacuum) is the state where all negative where D.Ep)D. In this perturbation expansion. then we easily realize that Eq. Pok^r^kka) XA(kko)d3kdk0.12).13) it is replaced by a /^independent quantity. 2. Thus taking particular plane waves ups*e~ivz+ip'>'.'e'?'. (2.17) .136 652 YOICHIRO NAMBU where S and P have the Fourier representation S(ppo) = ig*TSS*(p)8(po)h*(0)A(0) X f Tr^Gip'Po^pdpo ^ + ^ + +. For in the perturbation series the 2 obtained to any order is a function of pa.14) H0'= (e+x)?"3+<£n to up*S(p. take TrluGippo) XG(p+k. = iT3+<j>Tl. and expand 2 itself with respect to g . OND=(i/2)Tr(AOr2). whereas in Eq. we easily find that the diagonal matrix element of 2 corresponds to 2(p) = x(p)T3+<f>(p)Th Upt*S(p. As stated before. (2. Suppose we expand G occurring in Eq. 2. however. The corresponding Green's to be real. (2.. Hence there will be a correction left out in each order (analogous to the radiative correction after mass renormalization in quantum electrodynamics).12) corresponds to the ordinary Hartree potential which is just a constant. (2. It must be said that the Hartree approximation does not really sum the series of Fig. being a twocomponent wave function. A similar situation holds also for the photon selfenergy II. II must be identical with S. The first term in 5 of Eq. f~\ ~ig2f rsG(pk. we will use the trick 0D=Tr(AO).Ep. only one of which is occupied in the ground state. Eq. let us consider the meaning of Eq. The selfconsistency requirements (1. the selfconsistency rela. we have positive energy particles exist. FIG. In Eq.13) is always proportional to r 3 on the energy shell since Ho « TS. (s=l. Expansion of the selfconsistent selfenergy 2~S in terms of bare electron diagrams.12) whereas the second term gives an exchange effect.15) E=±Ep=±{ip2+4>f)i.13) in terms of perturbation Po+ipT3 + <l>pTl (2. G(ppo) = iptEJ+it ' with respect to 2 : G=GQ—ZGQ2GO—Go2Go2Go"f" • • *. can have in general two eigenfunctions up. the diagonal This form bears a resemblance to the Dirac equation. 2 In order to extract the diagonal and nondiagonal parts in spin space. however. ND signify the diagonal and nondiagonal energy "quasiparticles" (E<0) are occupied and no parts in the "spin" space.) function G now has the representation Before discussing the general solutions. (2. the Fermi surface. P (a): for the diagonal elements [on the energy shell.5) and (1.9)]. 2 completely since we equate in Eq.11).2) with different energies Ep. S in Eq. which are illustrated in Fig. (2. s^s'(p0'= —p0). (2. to note the possibility of a nonperturbational solution by assuming that 2 contains also a term proportional to TI or r 2 . the approximation is characterized by the fact that no phonon lines cross each other. Accordingly 2 will be < T% and commute with * Ho. 2(pEp)D=S(p.
As is well known. it is clear that TI<J> can actually be pointed in any direction in the 12 plane of the r space: r 101+^2. (2TT)4 J r I (epk<l>p{ep<f)pk) h{k)Vk x. not only on the energy malization of the phonon field. which should determine the renor.3 (2.22) • cv.~0). namely. 02=0. It has a solution <£~Q m exp(l/FiV).ko) in Eq.I t was thus sufficient to take </>i^0. (2. by extending the Hartree self.15).0) —> {<!> c o s a . In the last section we remarked that the selfconThe phonon selfenergy IT may be studied similarly sistency conditions Eq. (2.18) essential role in superconductivity.12).18) and rewrite it Finally we make a remark about the Coulomb inter U)=U+> (2 2i)  gir (• <pp_k h{k)d?k gir r <j>p—Ap{2TYJ EpkSlh(Ep. 3. the corresponding conditions of Bogoliubov antiparticle conjugation C of the quasiparticle field &. and (2. Any other solution is obtained by a transformation * — exp(iaT3/2)^. If a finite solution transformation with a constant phase. (See the following section.changes upspin and downspin electrons. the denominator represents the screening of the Coulomb interaction. NONLOCAL (ENERGYDEPENDENT) SELFCONSISTENCY CONDITIONS P *P I [ p~~ i (2*yJ I Epknk{Epk+ak) > This is essentially the energy gap equation of BCS if g*Aph(ky/tik(Epk+tik) is identified with the effective interaction potential V.18) is equivalent to. the consistency conditions to all virtual matrix elements. but slightly difFor later use. has a trivial solution <£=0.137 QUASIP ARTICLES IN SUPERCONDUCTIVITY 653 Applying this to Eq. Pk—<t>p<f>pk) g"ir tp(j>p= JEP?(EpkQk. C: L(p)*LC(p) = L(py. Eq. the phonon interaction factor gih(k)2A(k.2. N being the density of states: N=dn/dtp on the Fermi surface. (2. (2. (2. we finally obtain the following equations for x and <j> tpXp+4>v En g2T J EP Epk\$lk r YEP iij. (This is the same kind or of ambiguity as one encounters in the derivation of a potential from field theory. It does not play an shell (diagonal) and for the virtual pair creation out of . coming from the nondiagonal In view of the definition of ty. The difference between the local operator 2 and the nonlocal one S shows up in a situation like that in Fig. (2.13a) with Eqs._i+! \Slk FIG.) From the nature of Eq.This is defined by diagonal part of the selfenergy operator.19) eph(ky<Pk A = [fh(kyA{kh)+iei/kiy {ljn(^o)[A(^o)+ieVg2//®2F]}.20) is a gauge condition. 3. We will discuss this situation in a separate section since such a generalization brings simplification where T means transposition. in dealing with the problem of gauge invariance and As a consequence. Discussion about this point will be made later in connection with the plasma oscillations. > (2.p) between S and 5 is not complete.o nonlocal. —p).pA(Epk\Hk) E (2. The second equation. Equation (2.12).13). <f> sina). or interunpleasant situation. we have also collective excitations. The two j^\ 2 selfenergy parts overlap in time.20) (0. and if e p ~6 p (Xj.13) may be extended to all from Eq.14). (2. (2.) We may avoid this into holes of energymomentum (—p0. When this is taken into account.+qrc=Qqf+=Tly+i actually due to an inherent ambiguity in approximating nonlocal operators by local ones. this process will not be eliminated by the condition 2ND = (Si )ND' ' * A EpEpk h{kf X(«. we also mention here the particleferent from. Under C. and their centers \ of time h and h are such that li>fc. r operators transform as but this would mean that <j> (and x) must be treated as C: 1=1. reflection of the gauge invariance.k+Sit) (2TY J action. if F i V « l .virtual matrix elements. though it gives rise to an important correction when the Coulomb effect is taken into account. Thus the arbi<j> exists. An example of the situation where the cancellation of 2ND versus SND is not complete.12a) has to be replaced by (2.23) For the moment we consider the second equation of (2. and the compensation and changes quasiparticles of energymomentum (po. which is C: •q. If calculated ac\/J cording to the usual perturbation theory. 3. it cannot follow from perturbation treatment trariness in the direction of <f> is the 12 plane is a since there is no inhomogeneous term to start with. (2. because of a slightly different definition of the non.
1) i I TdG(p—k. (3.18) since we may just equate the coefficients of single particle may not exist. only an approximation to this sum. =E{pp0)yz{ppay\Pe=EAV)\ (3. x. 139 (1958) [translation: Soviet Phys.> ' ) A ( p . (3. our assumption about the analyiticity of G and II is consistent with Eq. Phys. Such a quantity is 13 It would seem then that we lose the advantage of the generalization since we cannot find the Bogoliubov transformation. parts (in the sense of Dyson) as an infinite sum of the and write p0Z(pp0) + e (pp0) T3+4>(ppa)Ti special class of diagrams illustrated in Fig. G(ppo) = i/L(pp0) In later calculations we shall encounter various integrals which we may classify into three types redx garding their sensitivity to the energy gap.is stable and the quasiparticles and phonons have a 14 less.7) (3. M.13) is imposed. In the following. <i> are even functions of pa. Theoret. Eqs. The X(pp0)=S(pp0). U. for example. B.6) may where f. 1. we get the energy Let us assume that these determine the approximate gap equation renormalized dispersion law 4>(p'po) *r p?=Er{py. 34.R.1) has a precise meaning in the bare particle valid physical meaning as excitations. M(k. i. Galizkii and A. as PoZ(ppoy—E(pp0f+ie was noted there. U(kko) = P(kko). 2. a o p<?—x+ie normal selfenergy part. Nuovo cimento 11. (3. kap0) system.p0)u^0. Migdal. we could still start from the older solution (2. etc. JETP 7.5) (2. . Exptl. (3.S. This simply means that 0 and IT are now nonlocal. namely. Such single particles will satisfy (2TT)4 J XTSG(pp0)ld°pdp0. (3. The earlier condition of Eq. G(ppo) — i .13) as the zeroth approximation to Eq. In particular. n on both sides.19).k0)=0This equation for 2 is much simpler than the previous We use the approximate equality since a really stable one (2. but also for the selfenergy effects which appear in intermediate states of any process. (2.1) are not exactly The Hartree equations now take the form identical even on the energy shell. It defines the (proper) selfenergy we will generally consider this quasiparticle peak only.7) The Green's functions G and A will be given by which implies that S and II are also analytic except for a cut on the real axis. The HartreeFock approximation based on Eq. and then calculate the correction. then be difficult to establish.13) represented. 342 (1954)] which can be derived by defining the Green's functions in terms of Heisenberg operators. However..6) the small fraction ~</>/£f of the electrons near the x2Z{pxyE{pxy Fermi surface in a superconductor. ko2=Ur(k)2.S. (3. could be interpreted as a nonperturbation approxi. 14 This is a representation of the Lehmann type [H. At any rate. These corrections would take account of the singleparticle transitions which remain after the Bogoliubov condition (2. Eq.1) imaginary part in the integrand is expected to have a 2 Actually.f t o o ) ] 2  P o 2 = ^(p) 2 not be much different from the older solution to Eq. and are to be completely equated with 5 and P.4) which is to be compared with Eq. See also V.13) and (3. the "radiative" correction to the Bogoliubov vacuum and the Bogoliubov quasiparticle.3) 4>(PP°) = (2iry / J p<?Z(p'poyE(p'p0'y+ie If we write for 2 2 X&(PP'. A(kk0) = i/M(kk„) dx •f •'n • Im x\i M(kx) This representation assumes that G(po)[_A(k0)2 is analyticexcept for a branch cut on the real axis.po—ki i)T3h(kk0)2d3kdko.138 654 YOICHIRO NAMBU the vacuum. (3. p0p0')dsP'dpo. (3. and is insensitive to the change of X Im . First. (3. Lehmann. these selfenergies can no more be incor. A > . Although the existence of a solution to Eq. should £ r 2 W = [ e ( ^ o ) 2 + 0 t o o ) 2 ] / [ l . (3.delta function or a sharp peak at x=E?{p) PM^ )]porated in H<s as the zeroth order Lagrangian since they These properties are necessary in order that the vacuum contain infinite orders of time derivatives. (2. perturbation theory. 96 (1958)]. (2. if it exists.2(ppo)(2TY J mation to determine the "dressed" single particles (together with the "dressed vacuum") or the Green's r r function (01 TQr (xl) . r3.19).6) or (3. the solution.p ' .8) 2ti>po) = poZ(j>po)+x(Ppo)Ti+<l>(ppo)r1.13 Neverthe.1). represents the effect of the bulk of the surrounding electrons on a poZ (px) + e (px) T3+4>(px) n particular electron. J. depend both on energy and momentum arbitrarily. In other words. (3. respectively.2) L(p.*+ (x'(')) 10) for the true interacting U(kk0)=i Tr [jsGikp.e.
1/E2.2) does not. arise from the rearrangement of yj/ and ^ + .1) dl dl c FIG. (VT3ieA)*(x)} . which Eq. Construction of the vertex part r in bare particle picture. since dP/dl+Vi = 2V+<f>T:?i'. and hence negligible (g 2 iV«l). For we have seen that the selfenergy 0 of a quasiparticle is a gaugedependent quantity. (4.2+02)i makes little contribution if fip—k) is a smooth function. 2. Theoret. such an expression is always of the order 1. etc. ¥(*)}). e p—>p—r 3 A (4. the energy gap itself is determined from an equation of the form g* r d3k —/(pk)~g2 J Ek d3k _/(p_k)~l •'&<<*„ Ek r . Koba. we will not be primarily concerned with the ordinary selfenergy effects. (4. the original Lagrangian £ has to be modified according to the rule d d e i > i—f. It can also be inferred from the gauge transformation ^—> exp(io.V . however. So the ordinary chargecurrent operator turns out to be in our form given by p(«) = ([*+(x) I T. The integrals are thus of the order g2N<j>/ccm. 4. being infinite C numbers.eA o. In the following. then it is clear that the electromagnetic current of a quasiparticle must contain. in addition to the normal terms given by Eq. It is well known16 in quantum electrodynamics that. acting on St'. . When we carry out perturbation type calculations.& A ) * + ( * ) .l). in the integrand which restricts the contribution to an energy interval ~2<£ near the Fermi surface.2) where the region ek<Ek = (ej. 10) which means that even if g2 is small. etc.{ ( r . (3. This expression.139 QUASIPARTICLES IN SUPERCONDUCTIVITY 655 sentation.11) Ek3 J Ek3 They have an extra cutoff factor ~ l / £ . The gaugeinvariant interaction of a quasiparticle with an electromagnetic potential should then be obtained by attaching a photon line at all possible places in the diagrams of Fig. p —> p — A dl dl c for the electron. (3. and will actually be compensated for by the first terms. In other words.. We will assume that proper renormalization has been carried out. and treat quantities of the third type as small. V .r3)'*' as was observed previously. Going to the twocomponent repre15 Hereafter we will often use the fourdimensional notation x= (x. 4. *(*)] + {*+(*). perturbation diagrams can be grouped into gaugeinvariant subsets. (Kyoto) 6. Second. (4. terms which would cause a physically unobservable transformation of <f> if the electromagnetic potential is replaced by the gradient of a scalar. If we want to have the quasiparticle picture and gauge invariance at the same time. this corresponds to the prescription d i d >i—1CT3A0.* ' e 7 V 0 ^ ( x ) . g W . In order to find such a conserving expression for charge current. The second line represents the polarization diagrams. 322 (1951). we will arrange things so that quantities of the second type are taken into account rigorously.2). in any process involving electromagnetic interaction. Progr. or else simply disregard it unless essential. Phys. INTEGRAL EQUATIONS FOR VERTEX PARTS16 / The second terms on the righthand side. 16 Z. The result is illustrated in Fig. p = (p. Finally we meet with integrals like €*</> f <j> —f(vk)d% £ —f(pk)d% etc. it is instructive to go back to the bare electron picture.9) '/ 4m + [ ( . in which the selfenergy is represented by a particular class of diagrams discussed in the previous sections. Such a subset can be constructed by letting each photon line in a diagram interact with a charge of all possible places along a chain of chargecarrying particle lines.*(»)]+{^(*) ) *(j!)}) 1 2 given by an integral like f(pk)d% (3./>o). In the presence of an electromagnetic potential. such that the invariance is maintained by each subset taken as a whole. dlp=d3pdp0. 4. has to be modified when we go to the quasiparticle picture. the complete charge current of a quasiparticle has to satisfy the continuity equation.
) dt/ \m ml W W XA'(p'p)+0(A*).p) = (b) yM(p'. (3.140 656 YOICHIRO NAMBU which consists of the "vertex" part r and the selfenergy part 2.6 Investigation of the higher order terms in A is beyond the scope of this paper.p) for a quasiparticle. 2.p) = TM(p'.+ T2L(P) = TW(p'. (4.p)) 17 will hold.p)=yi(p'. In this way we have shown the existence of spin and charge currents TiM(p'.3) is the basis of the rest of this paper. 1. \m ml (4. respectively: d dt d dt P m where y. Sir1" —> •*+er2<».7) in the presence of an external field. * + —»*+e r i a . The last two transformations of Eq. and we define 2«HP'. which do not depend on the presence or This equation may also be derived simply by considering the selfenergy equation (3.p) = (c) (d) 7 (4. ( 1 / 2 O T ) ( P + P ' ) ] which follows from Eq. These equations are identical with (4. so t h a t we obtain continuity equations for the z component of spin and charge.6) P m where W is the true Heisenberg operator. (4. = 32/3/1''.p)up = 0. we observe that there exist exact solutions to Eq.<») and r<a)'<6). I\—7. L(p')L(P) = up.p) = 0 The verification is straightforward by noting that G(p) = i/L(p). It expresses a clearcut approximation procedure in which the "free" chargecurrent operator y< of a quasiparticle is modified by a special class of "radiative corrections" due to Hint'. (4. which bear the same relations to 7 ' and YMAd) as Eq. T^(p'. Note that the above equations are unaffected by the presence of the phonon interaction.6) did to 7<«>. (4. but mix $1 and ^ j + in such a way as to keep ^rsJf invariant.2).p) = L(p')r1+r1L(p).7). In this way we are led to consider the modification of the vertex due to the phonon interaction in the same approximation as the selfenergy effect is included in the quasiparticle. (a) and (b) correspond. (4. respectively.P)=Z<J>W(P'P)+ 2 (W. i = 0 . M —> e T ! °*. * —> e " " * .*T^(p'. From infinitesimal transformations of these kinds we get ( \d* I*P+T ) w+v dtl • »r+7 2 ( — H — )*F = 0.1 ( \dl "W+V *+Tl ( . These solutions express continuity equations and other relations following from the four types of operations.4) «>(^) = W ) n + T i W ) . XG(pk)T3h(kyMk)d% pk) (4.*L(p') = 0. TW(p'. and expanding X in A.b) (4. to the spin rotation around t h e z axis. (4. T h e entire Lagrangian is invariant under them.are singleparticle wave functions satisfying L(p)up=up.7) will become giC"'") •*(\p'\W+(x)'Y("'>W(x)\p) Lo{p')Lo(p) = (po'—po) ~ r3(ep— ep). 3 stand for the free particle charge current [T 3 .3) absence of the interaction: (a) (b) (c) (d) •$(x)*eia(x)$r(x). * + —> •$r+eiT3a.p) = L0(p')T2+riL(P). Up')T3TzL0(p) = (p»'po)Ti(tp—ep).p) and Ti(b)(p'. uP. In the limit p'—p = 0.p)yi{p'. the lefthand side of Eq. It is not difficult to see that it corresponds to a "ladder approximation" for the vertex part.+ .p) = T^{p'. L{p')rzriL{p). and making use of Eq. Equation (4. a n d t h e gauge transformation.p)gi j'TGWKjTiip'k. The fact that we can find a conserved chargecurrent .W[2W2(?)]. which is the content of the Ward identity. (n=a.4) are not unitary.7) Taken between two "dressed" quasiparticle states. yW(p'.5) where a(x) is an arbitrary real function. The fact that there are simple solutions is not accidental. 2 should be now a function of initial and final momenta.p) = L(p')T. As the next important step.. for which the continuity equations (Po'po)T0Mt(p'p)rTi^ «—1 = T. and we get an integral equation17 Ti(p'. (4. > 3 •<Z+(x)>il!+(x)eia(x).3) for the following four types of vertex interactions (a) yM(p'.7') where up. Similar equations may be set up for any type of vertex interactions. * —> e" "^.
On the other hand. But then we have Km(g) — 0 as g—> 0 * since the intermediate pair formation is suppressed due to the finite energy gap. K.. KijV^Bijtif/m.explicitly. or deposits it with.6) where the properties of 7.4) easily. The diagram for the kernel K&K 'W r^. (5. we arrive at the conclusion that either one of the vertices 7 in Fig. and inserting the phonon interaction XT&(p))dlp 1 4 gi (2ir) m j TrirMpWP. 2TTY J where n is the number of electrons inside the Fermi sphere. this lack of gauge invariance should be remedied by taking account of the vertex corrections. establishing the unphysical nature of such a potential.4) Ji(q) = KijqMq) = 0.*•.2).C D  for a quasiparticle is rather surprising. 7.+ O* O . the surrounding medium which extends to infinity. This situation will be studied in Sec. S. continuously picks up charge from. . whereas Ka) corresponds to the diagram in Fig. Substituting this solution in Eq. (5. (5. the part Ka) is. 6. (4. We must conclude than that an accelerated wave packet of quasiparticles.¥(*)] 10)+ <*+(*). Graphical derivation of Eq. 6. Although we do not know T. whose energy is confined to a finite region of space. (5. In the case of a superconducting state.j<» is then Kii^{q)=—T Trlyi{pg/2.*(*)» (5. The thick lines represent quasiparticles. p) of Eq. yi (s. and to a hole well below. / » = Sij—«01 2m ^KijW+KijW. [**(*). however.2) takes the form of the London equation.5).OH.. as indicated in Fig. q)qi yi JibdlLKiteUte).and G under particle conjugation and a translation in p space were utilized in going from the first to the second line. In addition.(p. Thus Eq.p+q/2) X. where we will derive the charge current operators T.0 FIG. except that even a longitudinal field creates a current.5) we find Kiim(q)qi =— ( (2irY J Tr{yi{pq/2. 5. we have (5. For . (5. 5 will be replaced by quasiparticle lines. S. [Compare also Eq. For free electrons. According to our previous argument. (4. there is the polarization correction represented by a string of bubbles. (4. first neglect this correction.p)yi(p. (5. We calculate the Fourier component of the current J(q) induced in the superconducting ground state by an external vector potential A (q): XYAp+q/2.™.7) .141 QUASIPARTICLES IN SUPERCONDUCTIVITY 657 FIG.2) given by £ . GAUGE INVARIANCE IN THE MEISSNER EFFECT . A quasiparticle cannot be an eigenstate of charge since it is a linear combination of an electron and a hole. tending to an electron well above the Fermi surface. (4.4) where qn is equal to zero. 5 has to be replaced by the full T10.p+q/2)G{p+q/2) pq/2)G{pq/2)]d*p. (1.E ?t(P+q. (5. the free electron lines in Fig. we can establish Eq.p) explicitly.] It is well known that in this case Ka is of the form Kij(q)=(5irfqigi)K(q>).2) is exactly the solution Tm(p\q.i) where q is kept finite. Let us.lr&(p+q/2)G{pq/2)Ti2)d*p 1 [ Tr{[yi(pq. K(I) comes from the expectation value of the current operator Eq. effects. K is represented by K^KuM+Ki. whereas Ka) is essentially unaltered. according to Eq.5) (2ir) 4 J We will next discuss how the gauge invariance is maintained in the problem of the Meissner effect when the external magnetic field is static.3) p+q)~] so that for a longitudinal vector potential A «(j)~ JtX(g). Starting again from the free electron picture.1).
H. The calculation of K{1) from Eqs. people3 have shown already that the essential difference between the transversal and longitudinal vector potentials in inducing a current is due to the fact that the latter can excite collective motions of quasiparticle pairs. (Kyoto) 5. Fukuda and T.1) In other words T2<i>(p)=$o(p) satisfies a homogeneous integral equation: MP) = " W (5.— 4 f Tr[r3G^)>VTO (2TT) J (5. Theoret. (4.7) and (5.4). Although the proof is thus rigorous. Kinoshita. Actually there is a certain ambiguity in the evaluation of K<2\ Eq. Kilh) and if(2) are all infinite.J (2ir)f 3 <f p"Ep33ni •qid3pJrO(q2)qi^Ncrqi = n(el/m)qi. We see that the existence of such collective excitations follows naturally from our vertex solutions Eq. 6 TT)3 J 4E „ \ TO / (5. Progr. (5.MS/2)G(#+S/2) Xyt(p+q/2. The first term then gives '} e— 18 In order to understand the mechanism by which gauge invariance was restored in the calculation of the Meissner effect.8) (5. which does not depend on the quasiparticle picture. 1024 (1950). (The polarization correction is again zero. it is still somewhat disturbing since Kila). gives.. fcfc e) r 3 * (x. In this case we may write TM(p+q/2.t) ) 10} '*0 ± .12) We interpret this as describing a pair of a particle and an antiparticle interacting with each other to form a bound state with zero energy and momentum q=p'—p = 0. it is necessary to examine the collective excitations of the quasiparticles. and not as drastic as for the longitudinal case. and is just the anticommutator of the electron field.2). THE COLLECTIVE EXCITATIONS X{*+(x). we may say that the collective state saves gauge invariance.11') below. (5.6). (5.10) to show its formal gauge invariance since T3V— ieA(x) is certainly a gaugeinvariant combination for free electron field.2) /• 4? / p q \ 2 ( ) Pit.— Tr[r3G(xZ = 0)] = .<lPp * q%. (T3X>ieA(x))$r(x)) (5. Phys.10). and also to solve the integral equations for general vertex interactions. which is again similar to the one encountered in quantum electrodynamics. (5.1) using the bare quasiparticle states.11) and (6.p) becomes T<»(p.13) = «.3'19 It goes without saying that the effect of the vertex correction on Ka will be felt also for real magnetic field.2). (5. the way in which the collective mode accomplishes this end seems to differ from one paper to another. 6 and 7). 6. The second term also is finite and equal to 2m = . But as we shall see later.18 An alternative way would be to expand quantities in q without making translations in p space. The second term i £ a w comes from the cnumber term of the current operator (4. (6. pq/2)G(pq/2)dlp = 0. (1..p)=L(p)T3T3L(p) = 2JV 2 0.11) — f r3G(p'Mp')G(p') yj Xr3h(pP')2MPP')d4p(6.9) Thus [JT« 0 o ) (?)+^« ( i ) (?)]sy=0. / Thus the above proof is complete and independent of the Coulomb interaction which profoundly influences the polarization effect.J—<0[*+(*).T. the same value as Eq. as has been claimed by several people.pq/2) = i(p+q/2)i(pq/2) irfy>(P+q/2)+4>{pq/2)'] ^pq/m—2iT^>. For taking p = p'. (6. In fact. We will not attempt to analyze this situation here.) Since Eq. Therefore we may write for this contribution K<im(q)Mq) —ie = 1 giqxtfx 2m (2TT)3 • / The last line follows from Eqs. As for the polarization correction.5 . so that we get {Kiim+Kijm)qj=Q in the limit of small q.*(*)]0) Y0 ICH IRO NAMBU which is convergent and the same as the one obtained from Eq.SirTO i lim Z (01 T ( * + (*.13) is a contribution from the collective intermediate state (see Sees.142 658 The first term becomes further 8. 13 On the other hand. it is a small correction of order g2N (except for the renormalization effects). we can easily show in a"similar way that it vanishes for the static case (§0=0) because TrI\(jg/2. nor on the presence of interaction. . on the other hand. the second solution Yw(p'. .
a2~pF2/3m2. Salpeler and H. A.\q~p) = F^{p)+F^{p.p') This is an inhomogeneous integral equation for F™.q/2) = g 2 T—" 4 frsFV(P'.r~q2 =ccq. UV (p. j and the relative energymomentum (/»o. + f Tr F«» (p) u<x) (p. (6. 84.p).8).p) = Eq.q/2)L(p)+U«Kp.5) We then obtain From here on we carry out perturbation calculation. In other words.3) £ r L(p)F(p.9) AL(p. thus or F(\q+p. (6. thus. Rev. This condition can be derived as follows: We multiply Eq. We have to determine their dispersion law. (6.7) by F^{p)=G{p)^a{p)G{p). (6. Thus 1 p q.8) This is the desired condition. 1232 (1951).23).q/2)#p=0.q/2) XT3h(ppy&(pp')dipdip' In view of Eq. 100 (2TT) J Xr3h(pp'yA(pp')d*p'. Let us expand F and L in terms of the small change L(p±q/2)L(p). Bethe. we will neglect the p dependence of the selfenergy terms.q/2)d<p'.4) are the analog of the socalled BetheSalpeter equation20 for the bound pair of quasiparticles with zero total energymomentum. For a finite total energymomentum q. Collecting terms of the first order.(P)=L(iq+p)F(iq+p. (6.143 QUASIPARTICLES IN SUPERCONDUCTIVITY 659 In fact. Since there.7) (i)'(D'Ka' ™ where the average / is defined hjmtfp/$tfp.5) the last line is =f so that (77(0)^0))= C TrF<m(p)U<»(p. the inhomogeneous term U{p) must be orthogonal to the solution *o(^) of the homogeneous equation. and integrate thus: f Tr FV»(p)L(p)FV(p.q/2)L(p)d*p XrMpp')2Mpp')dlp'. there will also be pairs moving with finite momentum and kinetic energy.q/2)F™ (p)L(p) +L(p)FV»(p)AL(p. by defining F(p.)tF(#'. (6. (6. Equations (6. ±q/2). (2.q/2) = q0/2T3(p. The weight function <P/Ep*=<p/{.2) and (6.q/2)+ (6. Thus Ef^tf+tf. —p) is the fourdimensional wave function with the spin variables i. (6. 3 m2 .6) L(p±q/2) = L(p) + AL(p.! XHpp'y^pp'Wp.2) becomes G(p)Mp)G(p).iqp)L(phq) !ir)4 J J f [TrFW(p)TSFV(p'.q/2)dip 1 — ( 2x))4 T 4: f £ M«(r. the homogeneous integral equation takes the form *. F<» (p) = T*j>/(p02EJ+ie). e p ~ 0 .q/2) = AL(p. p)n = —— (2TT)4 J r3F(p'. For the evaluation of Eq. 20 q/2). there will be a continuum of pair states with energies going up from zero. 2 Tr L(p')F«»(p')L(P')FV(p'. Phys.11) K.q/2) (6. In order that it has a solution. we get L(p)FV(P.4) The particleconjugate quantity Lc was defined in Eq. E. p')ii J J. ( 6 . (6. so that p2~pF2.t *+<?)* peaks around the Fermi momentum. (6.p)L(p) or Z Lip^L^p^Fip. Fij(p. exists a bound pair of zero momentum. = g 2 — (2TY J [rMq+p'Aqp') XTMPP'YKPP'WP.q/2m+ (?/2) /2m).
This can be understood in the following way.1) for To.11). we find that all the terms are of the type (3. incidentally.2). the standard approach to solve the equation would be the perturbation expansion in powers of g2. (7.2) which can readily be verified. (i=0. which in fact agrees with Eq. =LZ(p+q/2)+Z(pq/2)}/2 p~q/2)L(pq/2) to the extent that terms of order g2N and/or the ^dependence of the renormalization constants are neglected. p'q/2) XrMppyA(pp')d*p'. In quite a similar way the current vertex r may be constructed. In other words. (7. p'q/2) xGip'qmrMppyMppw. we find that i? ± must be a solution of the homogeneous equation. so that r will contain terms of the form R±/(qo±aq). if we assume q=0.q/2)L(fiq/2) is even simpler. but does not take into account the annihilationcreation process of the pair due to the same interaction. to the zeroth order we may neglect the integral entirely and so FV(p. We get *. ^ ( ^ ^ ^ o ^ ) in the zeroth order in g2N.pq/2) under the condition qo±aq=0. Since corrections to the noncollective part of To also turn out to be calculable by perturbation. that there are lowlying collective excitations. we would find in each order terms of order 1.7).13) (6. we have still to solve the integral equation (6.11') cPN « pF /3irm — n. that 1 r<j?  {2wyJ E3 d —Pp^N^mpF/T?. (6. pq/2)/q0 = T3{ZZ(p+q/2)+Z(j>q/2)}/2 + (po/q°)LZ(p+q/2)Z(pq/2n} lx(p+q/2)X(pq/2)yq0 +irfy{p+q/2)+4>{pq/2)yqo.5) = r 3 f JrsG(p'+q/2)T0(p'+q/2. (7. The residues i? ± can be obtained by taking the limit R±= lim To(p+q/2. (6. (4. These processes can go through the collective intermediate state with the dispersion law (6.(0«*o(tf to this order. Z pq/2)~r3Z+2iTrtq0/(qo2a>y). (7. (7. We will consider it in a later section. Namely. The original function *t(P)=L(p+q/2)F(p.2) if C + = — 2%. qoT^O. (7. i?±=C ± * 5 (/>). and these excitations do not follow from perturbation. It is well known that this annihilationcreation process is very important in the case of the Coulomb interaction. 3) from their integral equations.q/2) = G(p)UV(p. To contains matrix elements for creation or annihilation of a pair out of the vacuum. This time we start from the longitudinal 21 If we proceeded by perturbation theory. or L(p+q/2)F0(p+q/2.1) in terms of T(6> of Eq.12) We know. t^l<t>(p+q/2)+4(pq/2)l/2.144 660 YOICHIRO NAMBU which is the dispersion law for the collective excitations.11). discussed before. 1.pq/2)(q0d=aq).4).3) Applying this procedure to the integral equation (7. pq/2) = T^(p+q/2. CALCULATION OF THE CHARGECURRENT VERTEX FUNCTIONS In this section we determine explicitly the chargecurrent vertex functions I\. 2.2'3 We also note. T0(p+q/2. The second term is the result of the coupling of T% to the collective mode.21 Fortunately. We would like to emphasize here that these collective excitations are based on Eq. For the particular case ? = 0 . goiag—*0 (7.4) 7. we may now put Toip+q/2. As for the wave function F ( 1 ) itself. which takes account of the phononCoulomb scattering of the quasiparticle pairs. however. then we have an exact solution to Eq. and plays the role of creating the plasma mode of collective oscillations. = T3+g Jr3F0(p'+q/2. iQ(p) reduces to T#f>(p). But this can be done by perturbation because on substituting J7(1) in the integrand. Let us first go back to the integral equation for To generated by r 3 : T0(p+q/2. 2 This observation enables us to write down F 0 for q ?^0. to which T 3 can be coupled.q/2)G(p). namely. (6. 6. According to the results of Sec. . Only the particular combination r ( 6 ) of these was given before.1) For small g2.
24)]. When the particle is accelerated. pkMpk^k TrtT3G(p'k) a 2 U j(*. for exXT(p'k.7) A(p'.3) for T with the inhomogeneous term replaced by y+r3DX.t)^e^T3Z^(x.2) Eq. (8. (2.4) .q/ff) F + 2 i r 2 < ^ V ( ? o 2 .a Y ) .11) X(p'p)=iJTrlT3G(p'k) XT(p'k./) = joV/1 where / satisfies the wave equation +iD(p'p)X(p'p)j l d / *\ XTo(p'k.l) P(x.10) for r. Putting X{p'p)=i f TilnG(p'k)A(p'k. XToip'^p^Gipk^kl 22 There may be transverse collective excitations (Bogoliubov.. reference 2). D(q)=igVl(qyA(q)+e*/q\ pk)G(pk)2d% •1) which is indeed zero on the energy shell. p^Gipk^k ample. but they do not automatically follow from the selfenergy equation nor affect the energy gap structure. j ^ O . whereas the collective part is spread out both in space and time. which now satisfy the following type of integral equations (7.have a very interesting structure. so that A is a linear combination of the T corresponding to 7 and To : A=r+ry?x.(p+q/2)+<t>(pq/2) — 2XTT The inclusion of the annihilationcreation processes in the equations of the previous sections means that the vertex parts get multiplied by a string of closed loops. Neglecting the momentum dependence of Z.6) and (7. =X(p'p)/llD(p'p)X0(p'p)2. which has the exact solution T(p+q/2. (7. We will call the new quantities A. pk)G(pk)Jd% (A )/=2e*+T 2 tf>tf.8).22 We may.12) or V a2 dP/ (po. (7.pq/2)q/q _ « (P.qY=qar3Z+ (p • q/m) Y+2iTi4> +W(p'p)riJ Tllr3G(p'k) XA(p'k.)£*t(p/m)F*(a. a fraction of the charge is exchanged between the core and the cloud. (4.pq/2)q/q=T^(p+q/2.p+q/2)/q _Pqj mq{ x(p+q/2)x(pq/2)\ p • q/m J r»po[£(p+q/2)!. then. The latter is surrounded by a cloud of the excitation field / ..6) and (7. The noncoUective part is essentially the same as the charge current for a free quasiparticle except for the renormalization Z and Y. therefore.« Y ) . / will fall off like 1/r from the core. we may thus write the chargecurrent density (p. (78) Y= l+lx(P+q/2)x(P~q/2)y(p<l/m).1) takes the same form as Eq. Combining (7. i . which represent the polarization (or shielding effect) of the surrounding medium. on the other hand.l)\ 2 D(q) includes the effect of the Coulomb interaction [see Eq.. (7.(pq/2)yq 4.2) then yields (8. (8. The transversal part of r . xlliD(p'p) f TT[_r3G(p'k) The total charge residing in a finite volume around a core is not constant because the current — V/ reaches out to infinity.jo) is the chargecurrent residing in the "core" X(p'p)=if Tr[T3G(p'k) of a quasiparticle. is not coupled with the collective mode because the latter is a scalar wave.pg/2)~(Wm)Y + 2 i r 2 ^ a V ( 5 o .()v/(x. In a static situation. (7. (8.j) as 1 df{x. and <> /.3) 1 df = Po\ a2 dt .pk) XG(pk)r3D(k)d*k Iq For qa 5*0.10) Equations (7.p)=yifriG(p'k)A(p'k.145 QUASIPARTICLES IN SUPERCONDUCTIVITY 8.8) r(p+q/2. Substitution in E q . write instead of Eq. p'k) XG(pk)2d% (8. the continuity equation takes the form q0rB. THE PLASMA OSCILLATIONS 661 component for g 0 =0. we get r(p+q/2.
13) = if Tr[. (8.11). and q0^>4> or <K0. We would like to know. has indeed the correct property in spite of the fact that the " b a r e " vacuum.7) X0(q. (8. (8.qo) • « \qo—En qo+En/ IC0) 2 .11) excitations. mation by including processes (or diagrams) which We see thus that the previous collective state with have not been considered here. Xo should vanish for 9—> 0.T3G(p'+q/2) X9q(p')G(p'q/2)r33dY. the corresponding approxiay/4/1 mation insures. $ 0 ^ 0 since the righthand side then consists of the nondiagonal matrix elements of the total charge operator Q: Xo(0. where n is the number of electrons per unit volume. « ^ 0 for some 50?^ 0. (8. We can also study the behavior of Ao in the limit go — 0 for small but finite j 2 : > A. (8.8) where X0(q) is defined in Eq. hence should be neglected to be consistent with our approximation. This is posr 2 8 10 Xo^Wi\ /(?o aY)( . the dominant part of D{q) in Eq. we obtain 1 X0 = f aV r <f>2d3p If 0) is an eigenstate of charge. QM = ifr^(p'+q/2)et(p') XG{p'q/2)rMp~P')dip' +iTj)(q)f which means Qt(p) = T0(p+q/2. if treated properly.6) pq/2)D(q)X(q). E? # 3.) sible because we are taking account of the "radiative For small §2. After some simplifications using Eq.8) corrections" to the bare quasiparticles which are not is the Coulomb interaction e2/q2. CONCLUDING REMARKS <trd p We have discussed here formal mathematical struc+• ture of the BCSBogoliubov theory.146 662 Especially for Y = T 3 .p) = r0(p'. There are some questions which have been left out. approximation is characterized essentially as the HartreeFock method. (8. from which we started. we get X0(p'p)=Xo(p'p)/£lD(p'p)X0(p'pn Ao(p'. as is clear from E q s . the second integral may invariance is maintained. X(q) Tr lrsG(P'+q/2)eq(p') XG(p'q/2)y*p'. (8. I n t h e E„ presence of external fields. (8.8) can be calculated. (8. the quantity X0 in Eq.4). Our result for Xo. but again agrees with the free electron value. The converse is also true if En> ?o.a V / 4 ) G —)} .9) pretation in terms of perturbation expansion. Eq.(g. The solutions to Eq.qo)= f {0\T(p(xi). 3 (2wY L(7o2aV J £ p ( £ / . is of the order g2N.p)/llD(p'p)X<)(p'. and can be given a simple inter+0(qi). Xo represents the charge density correlation in the ground state: —f dzp~N.9). (8.5) and (8. let us next write down the homogeneous integral equation: YOICHIRO NAMBU On the other hand.pn To obtain the collective excitations.qo) is the following. Compare Anderson. With the solution (7. reference 7. for one thing. if Coulomb interaction is neglected.8) leads to23 ?„ 2 =ay[l7g 2 A( ?1? „)/. (2T)3 J which comes entirely from the noncollective part of To. 9. Another observation we can make regarding X0(q.8) determine the new dispersion law qo=f(q) for the collective excitations. we s e t l = D(q)Xo(q). (8. Substituting @e in the second equation from the first. what will happen This agrees with the ordinary plasma frequency for free if we seek corrections to our HartreeFock approxielectron gas.p(0))\0)e^+i'">tdhdt. is not an eigenstate of charge. (6. T h e nature of the 2 2 AJ / Ep(qo /4EJ)\3m EJ .6). Even within our apg02=a2q2 has shifted its energy to the plasma energy as 23 a result of the Coulomb interaction.12) The correction term.?o) 2 A 7 ]. however.8) then eigenstates of charge. (8. Equation (8. These corrections manifest thembecomes selves primarily through the existence of collective ?c 2 =eVW=e 2 » (g 2 ^0). I t is interesting that the be dropped and quasiparticle picture and charge conservation (or gauge invariance) can be reconciled at all. that the gauge For aq<&j>.
25 K. Androes. Letters 2. F. Yosida. . 769 (1958). Phys. 24 The collective excitations do not play an important role here as they are not excited by spin density. G. M. Androes and VV. Phys. Phys. ACKNOWLEDGMENT We wish to thank Dr.4). Rev. there is an additional assumption of the weak coupling (gW<5Cl).147 QUASIPARTICLES IN S U P E R C O N D U C T I V I T Y 663 proximation. Experimentally. R. Rev. 852 (1956). 386 (1959). 208 (1957). 104. Rev. and the importance of the neglected terms (of order g2N and higher) is not known. Eq. [r < o ) .25 Knight. 106. (4. Reif. Schrieffer for extremely helpful discussions throughout the entire course of the work. Rev. D. Knight. this does not seem to give a finite spin paramagnetism at 0°K. 110. Phys. and Hammond. This effect has to do with the spin density induced by a magnetic field and can be derived by means of an appropriate vertex solution.3 It is desirable that both experiment and theory about spin paramagnetism be developed further since this may be a crucial test of the fundamental ideas underlying the BCS theory. does not have the characteristic pole. However. there has been some evidence24 regarding the presence of spin paramagnetism in superconductors.
This behaviour is of fundamental importance to the argument that follows. of the collective excitations of zero energy 3) results in an unphysical restriction in the free choice of phases. equaltime anticommutation relations and number operator relations may be derived from those of the Bogoliubov theory by requiring both sides of the equations to have the same effect on JV^ and Nr. is to treat the term in the Kamiltonian which transfers electrons across the barrier as a perturbation. However. ^).N. due to the fact that we have a system containing two disjoint superconducting r e gions. Relations expressing electron operators in terms of quasiparticle operators. where Hj< expressed in electron operators is t These are equivalent to the operators which change \N) to [V + 2) in the theory of Gor'kov 4 ). we take our unperturbed Hamiltonian to be H0 = Z nk Ek + \[ Ni + \ r Nr . as in the phenomena dealt with here. The S operators. and the Department of Scientific and Industrial Research. corresponding to the Bogoliubov operators a£ we use quasiparticle creation operators a e ^. A difficulty. but may be avoided by working with the projected states with definite numbers of electrons on both sides of the barrier.XN (x = chemical potential). following that of Cohen et al. but they can depend on the relative phases of the S operators associated with two superconducting regions. the numbers of electrons on the two sides of the barrier.148 Volume 1. a s n ek %k = k kNoting that the Bogoliubov Hamiltonian is H . Bogoliubov et al. if the regions are independent these states must be capable of superposition with arbitrary phases. and derive the commutation relations [•^o^efe] = (Ek + x k) aek > [H0'Sk] = 2 Xk Sk • In the presence of tunnelling the Hamiltonian is H0 + Hp. The phase of an S operator is related to the orientation of the plane containing the pseudospin operators 3). N. a ^ which respectively add or remove an electron from the same side as the quasiparticle and leave the * Work supported by Trinity College. and we normalise them to have unit amplitude. England Received 8 June 1962 TUNNELLING * We here present an approach to the calculation of tunnelling currents between two metals that is sufficiently general to deal with the case when both metals are superconducting. and pair creation operators S^ which add a pair of electrons on one side leaving the quasiparticle distribution unchanged. The Hermitean conjugate destruction operators have similar definitions. Our procedure. in the quasiparticle approximation. 4 t = uk aefeO + vk a h£l ' a ek aek =n k ' . number 7 PHYSICS LETTERS 1 July 1962 POSSIBLE NEW E F F E C T S IN S U P E R C O N D U C T I V E B.D. ** W shall use subscripts I and r to distinguish operae tors on the two sides. and k to denote an operator referring to either side. J t t Cf. The neglect. In that case new effects are predicted. number on the other side unchanged. whose corresponding number operators are constant. JOSEPHSON Cavendish Laboratory. 251 . In particular. k where Ek is the quasiparticle energy in the Bogoliubov theory. On switchingon the transfer term the particular phases chosen will affect the predicted tunnelling current. referring to macroscopically occupied states. *•'. Physical observables cannot depend on the phase of a single S operator. may be treated as time dependent cnumbers . This is because states defined as eigenfunctions of the Bogoliubov quasiparticle number operators contain phasecoherent superpositions of states with the same total number of electrons but different numbers in the two regions. Cambridge. Corresponding to these projections we use operators which alter the numbers of electrons on the two sides by definite numbers **. arises if we try to describe quasiparticles by the usual Bogoliubov operators 2). For example. Cambridge. due to the possibility that electron pairs may tunnel through the barrier leaving the quasiparticle distribution unchanged. We assume that in the absence of the transfer term there exist quasiparticle operators of definite energies.
Er ^eV . r If we describe the time dependence of operators by the interaction picture °>.  JjJ is then equal to the current flowing in the normal state at an applied voltage equal to n times the energy gap. (ii) is due to pair transfer without photon emission.nrl) (ii) at zero voltages. and by r e taining only those terms in Jjjit) which can be expressed in accordance with (1) as products of S and number operators obtain an expression equivalent to the usual one. and currents in the films destroy the timereversal symmetry and reduce \J\\.nlo .i / / z rt eet' HT(t<) &t')} . The current operator in the Heisenberg picture is related to that in the interaction picture according to Jtft) = U~Ht) JrntW U(t) . At low temperatures and voltages the first term of (4) dominates. The remaining terms oscillate with frequency v = 2eV/h (V = Xj . and at high frequencies the effects are reduced by the capacitance across the barrier. the phases of the supercurrents will vary rapidly over the barrier. equations (2) imply that the a and S operators have exponential time dependence. and typical term s are of the form 2ieulvlur j r J J Tlr T_h. Applied r. Hence the DC characteristic has a zero slope resistance part over a range of current dependent on the magnitude of the AC voltage. 1). t r the effective superconducting wave functions 7) in the films on the two sides. The second and fourth terms result from processes involving real intermediate states. and can be regarded as fluctuations in the normal current due to coherence effects.))] (4) where k denotes the state paired with k. number 7 f*r(Tlra+l a + T PHYSICS LETTERS rla+ral> • 1 July lj.v corresponds to 483. where j is the tunnelling current density. At higher temperatures the third term reduces J^  .f. where n is an in teger. Magnetic fields. 2eV/h = nv.r + [(1 . If for some n. a s sumed to be the same on both sides. and reduces for the same reasons to the usual one obtained by neglecting coherence factors. From (3) our theory predicts that (i) At finite voltages the usual DC current occurs. owing to the time dependence of the S operators. The possibility of observing these effects depends on the value of JjJ .149 Volume 1.£ r)3 + "16(e V + E{ . In larger fields. of the form H = ^0 + 2 JX sl S r + i J\ Sr Si . Equivalent quantummechanical explanations of these effects can be given. Consequently the photon frequency is not broadened by the finite quasiparticle lifetimes occurring in real superconductors. Anderson 8) has suggested that the absence of tunnelling supercurrents in most experiments hitherto performed may be due * [P'^}rTr " (»b " «„)) IP eV^El. J0 is similar to the expression of Cohen et al.7'i *f V +ij{ <ly */ . We note that the first term remains finite at zero temperature and zero applied voltage. leaving the quasiparticle distribution unchanged. causing the maximum total supercurrent to drop off rapidly with increasing field. This formula predicts that in very weak fields diamagnetic currents will screen the field from the space between the films. but a DC supercurrent of up to a maximum of  J±  can occur. J0 is zero. J\ is proportional to the effective matrix element for the transfer of electron pairs across the barrier without affecting the quasiparticle distribution.El .n. where U{t) = lim { T e x p ( . We also express lnt(0 = i e A [HT. (3) To second order in Hr. screening will not occur.Xr being the applied voltage). The linear dependence of the current on the matrix element is due to the fact that the process involves macroscopically occupied states between which phase relationships can occur.. Here Hp is expressed in the interaction picture and U(t) can be evaluated by writing HT in terms of qua siparticle operators and using the method of Goldstone 6).£.. but there is also an AC supercurrent of amplitude  JjJ and frequency 2 eV/h (1 p. but with a large penetration depth owing to the smallness of j j . Thus if a DC voltage V on which is superimposed an AC voltage of frequency v is applied across the b a r r i e r . The effects may be taken into account approximately by replacing (3) by J =. owing to the existence of a critical current density.7'0 + i. the supercurrent has a DC component dependent on the magnitude and phase of the AC voltage. and in the presence of timereversal symmetry all contributions to it are in phase. fields can be treated by noting that the oscillations in V frequencymodulate the supercurrent. For example (i) is due to the transfer of an electron pair across the barrier with photon emission. 252 .**r] 1} in terms of quasiparticle operators. and .6 Mc/s). the current has Fourier components at frequencies 2eV/h±nv./.
Gor'kov. JETP 9 (1959) 1364. 1959).1 gauss. 3) P. New York. Theoret.N. Exptl. 2) N. Cancellation of supercurrents would start to occur when the amount of flux between the films. (London) A 239 (1957) 267.Shirkov. 8) P.Anderson.Goldstone. V. number 7 PHYSICS LETTERS References 1) M. Phys.P. Soo. Falicov and J.. JETP 7 (1958) 505. Inc.Bogoliubov. Rev. effects similar to those considered here should occur and may be relevant to the theory of the intermediate state. Anderson and Prof. V. B. J.Anderson. Such a field would not be appreciably .Tolmaohev and D.W. (USSR) 34 (1958) 58. 6.Gor'kov.W. ** 253 . A new method in the theory of superconductivity (Consultants Bureau. When two superconducting regions are separated by a thin normal region. Exptl. translation: Soviet Phys. H. private discussion. J. P. L.N. 4) L. This would occur for typical films in a field of about 0. 112 (1958) 1900. translation: Soviet Phys. V. Phys. translation: Soviet Phys. 6) J. J. Rev. including that in the penetration regions. Phys. Roy. Phys. became of the order of a quantum of flux hc/2e. (USSR) 34 (1958) 735. (USSR) 36 (1959) 1918.excluded by the critical currents obtainable in specimens of all but the highest conductivity.P. Theoret. 1 July 1962 to the earth's field acting in this way. Exptl. Theoret. Phys. Cohen. 5) N. Phillips. JETP 7 (1958) 41. C. Letters 8 (1962) 316.Bogoliubov. 7) L. I am indebted to Dr. Pippard for stimulating discussions. W. A. p. Proc.150 Volume 1. M.
NUMBER 1 JANUARY Gauge Invariance and Mass JOLIAN SCHWINGER Harvard University. N. in strict analogy with electric charge. one should distinguish sharply between punerical gauge functions and operator gauge functions. California (Received July 20.nlt){np)+pltp. and will generally lack the positiveness properties of the radiation gauge. In each coordinate frame there P * unique operator gauge. the Green's functions. which can be derived from the unordered product (A. W Tw P le 'd J. 115.f). Rev. if m0 is exactly zero the commutation idations. once the assumption of weak coupling is removed. Massachusetts. B(m2) = Bv&(mt)+B1(mi). Los Angeles.2 may be open for the first time. D Green's functions of other gauges have more complicated operator realizations. Cambridge. OES the requirement of gauge invariance for a vector field coupled to a dynamical current imply the existence of a corresponding particle with zero gss? Although the answer to this question is invariably gi v e n m ^ e affinnative.151 PHYSICAL REVIEW VOLUME 12 S .. hPhys. in virtue of the gaugedependent asymmetry between space and time (the time axis is specified by the unit vector wM): I" A.9 75.(«).B{m*). but leaves unspecified a possible delta function contribution at m=0. or equivalent properties. pPhys.Schwinger.N . 721 /"mem •achwmger. n TIT**na andc C. The latter determines B(m2) for w > 0 . Indeed. it is easily shown3 that the mass spectrum Jnust extend below mo. Let us consider the simplest Green's function associated with the field A. The requirement of nonnegativeness for the matrix A^. The discontinuous change of invariance properties produces a corresponding discontinuity of the dynamical degrees of freedom and of the operator commutation relations.(x')) {dp) • / 0)! •eM^dm1 rt+ipWf+m^A^p).Yan s> v s . »T '•r •T i v " 5'hnson. No reliable conclusions about the mass spectrum of a gaugeinvariant system can be drawn from the properties of an apparently neighboring system.*(p) = B{nt) s 2 {p„n. where the factor ri+(p)8(p*\m2) enforces the spectral restriction to states with mass m > 0 and positive energy. A gaugeinvariant system is not the continuous limit of one that fails to admit such an arbitrary function transformation group. Yang.s o— l i e 7ii (1959). without the obvious conflict with experience that a massless particle entails. 1961) It is argued that the gauge invariance of a vector field does not necessarily imply zero mass for an associated particle if the current vector coupling is sufficiently strong.p. j„{p) = «2B(m2) (p. One potential source of error should be recognized at the outset. I t obeys the sum rule 1= J dm.RRev. Thus the path to an understanding of nucleonic (baryonic) charge conservation as an aspect of a gauge invariance. 1=50+/ Jo 'o drrPBxW).g„. for which f8* has the standard operator construction in a vector J*W of positive norm. with a physical probability fikfpretation.v . J* scnwinffp. if one considers a vector field coupled to a divergenceless current. characterized by threePjOensional transversality (radiation gauge). for the various operator gauges are not on the *>ne quantum footing.therefore become arbitrarily small as nto approaches t]Kro. The field equations supply the analogous construction for the vacuum expectation value of current products OVWi'C*'))) m terms of the nonnegative matrix If invariance under arbitrary gauge transformations J asserted.\ I "' f+(npY J Here £(m ) is a real nonnegative number. The nonnegative constant Bo is then fixed by the sum rule.Phys.{p) is satisfied by the structure associated with the radiation gauge. where gauge hvariance is destroyed by a socalled mass term with parameter mc. Nuclear Phys.W " M O " ' 1 . 25.4 The . eRev. 651 (1949).1 the author has become convinced that there is no such necessary implication. T>k. with a smaller inwriance group. The factor nfi has the decisive consequence that m=Q is not contained in the current vector's spectrum of vacuum fluctuations. When the theory is formulated with the E~ °* vacuum expectation values of timeordered S ^ t o r products. • Lee so m 397 .> 1 1501 (1955). the freedom t«*^aJ_gauge transformation can be restored. The lowest mass value will ..+p. however.. This situation may permit a deeper understanding of nucleonic charge conservation as a manifestation of a gauge invariance. Nevertheless. upon which this conclusion is based become entirely different and the ligument fails. and University of California.. 435 (1961). Jo which is a full expression of all the fundamental equaltime commutation relations.(x)A.8 98.
One.RT=~l. • Note added in proof.jLi and odd parity. H. We have now recognized that the vacuum fluctuations of the vector A „ are composed of two parts. Stevenson. that B(m?) shows a more or less pronounced maximum which could be characterized approximately as an unstable particle.152 39S JULIAN SCHWINGER erned by all of the dynamical properties of the fields that contribute to this current. of course. and M. at which point J50 = O. with w > 0 . in the approximate stronginteraction framework is that of the states with N=Y=T=0. is directly related to corresponding current fluctuations. Or it may be that the electrodynamic coupling is quite considerable and gives rise to a small value of Bo.and fourdimensional senses and has no accompanying current. Letters 7. A. ' The very short range of the resulting nuclear interaction together with the qualitative inference that like nucleonic charget are thereby repelled suggests that the vector field which definei nucleonic charge is also the ultimate instrument of nuclear stability. W. we have jBi(m2) = 0 and 5 0 = l or. C. which may be the situation for the electromagnetic field. L. and the relationship between these fundamental fields and the phenomenological particles can be comparatively remote. 178 (1961). and w = 0 disappears from the spectrum of A?? The general requirement of gauge invariance no longer seems to dispose of this essentially dynamical question. just the radiation field. while the other part. MagliJ. Experimental evidence for an unstable article of this type has recently been announced by B. Can we increase further the magnitude of the variable parameter until fdrr? B\(nP) attains its limiting value of unity. unit spin particle of nonzero mass? Not necessarily. i. If this is set equal to zero. Imagine that the current vector contains a variable numerical factor. Alvarez. For a sufficiently small nonzero value of the parameter. Rev. since the vacuum fluctuation spectrum of A i. in Phyi. Rosenfeld.6 But the essential point is embodied in the view that the observed physical world is the outcome of the dynamical play among underlying primary fields. Would the absence of a massless particle imply the existence of a stable. For the particularly interesting situation of a vector field that is coupled to the current of nucleonic charge. the relevant spectrum. becomes identical with that of / w which is gov E .6 It is entirely possible. can be associated with a pure radiation field. in contrast to the immediate correlation that is commonly assumed. This is a continuum. with m = 0 . beginning at three pion masses. which has the appearance of a fairly weak coupling. Bo will be slightly less than unity. which is transverse in both three.
ANDERSON Bell Telephone Laboratories.u1). Y. Pines. Phys.153 PHYSICAL REVIEW VOLUME 110. 125. three modes exist. Schwinger comments t h a t the commutation relations for the gauge field A give us one sum rule for the vacuum fluctuations of A. the two sum rules are normally incompatible unless there is a contribution to the A rule from a free. R is equivalent to the mass. however. (N. and Mass P.u}) 4 = <iirjli(k. massless solution of the field equations. show t h a t the q u a n t u m nature of the gauge field is irrelevant. For instance. T h e usual theory of the plasmon does not treat the electromagnetic field quantummechanically or discuss vacuum fluctuations. T h e plasma frequency . W.J. P. for instance) does not necessarily have zero mass. NUMBER 1 1 APRIL 1963 Plasnions. 397 (1962). Phys. but that the example of superconductivity illustrates that the physical spectrum need not. Since j is the source for A and the two are connected by field equations. a i. to denote any generalized gauge field accompanying a local conservation law. h Schwinger. yet there is a close relationship between the two arguments. therefore. Rev. weakly interacting. Sakurai. Rev. in close analogy (except for problems of Galilean invariance implied b y the inequivalent dispersion of longiuainal and transverse modes) with the massive vector { ^ o n j n e n t i o n e d by Schwinger. At and above this frequency. 741 (1958). I n the plasma. N. 109. and we. Our argument is as follows: T h e equation for the electromagnetic field is fAli=(k*rf)All(k. Rev. Nozieres and D. It is also shown that Schwinger's criterion that the vector field m?^0 implies that the matter spectrum before including the YangMills interaction contains m = 0. 1501 (1955). Phys. In spite of the absence of lowfrequency photons. Yang. while those for the m a t t e r field give a completely independent value for the fluctuations of matter current j . transverse electromagnetic waves do not propagate below the "plasma frequency. I n fact. the source term is large enough. Gauge Invariance. Schwinger 1 has given an argument strongly suggesting t h a t associating a gauge transformation with a local conservation law does not necessarily require the existence of a zeromass vector boson. one can draw a direct parallel between the dielectric constant treatment of plasmon theory 4 and Schwinger's argument. Murray Rill. 3 (We use the term "YangMills" in Sakurai's sense. J. it had previously seemed impossible to describe the conservation of baryons in such a manner because of the absence of a zeromass boson and of the accompanying longrange forces.) The purpose of this article is to point out t h a t the familiar plasmon theory of the freeelectron gas exemplifies Schwinger's theory in a very straightforward manner. 98. if a certain criterion on the vacuum fluctuations of the generalized current is satisfied. Some comments on the relationship between these ideas and the zeromass difficulty in theories with broken symmetries are given. Ann. E C E N T L Y . We show that the theory of plasma oscillations is a simple nonrelativistic example exhibiting all of the features of Schwinger's idea. If.) 11. 2 T h e problem of the mass of the bosons represents the major stumbling block in Sakurai's a t t e m p t to treat the dynamics of strongly interacting particles in terms of the YangMills gauge fields which seem to be required to accompany the known conserved currents of baryon number and hypercharge. gauge invariance and particle conservation are clearly satisfied in the plasma. 1 (1961). there can be no such contribution and the massless solutions cannot exist. while the finite density of electrons leading to divergent " v a c u u m " current fluctuations resembles the strong renormalized coupling of Schwinger's theory. . New Jersey (Received 8 November 1962) Schwinger has pointed out that the YangMills vector boson implied by associating a generalized gauge transformation with a conservation law (of baryonic charge." which is usually thought of as the frequency of longwavelength longitudinal oscillation of the electron gas. Phys. homogeneous.PLee and C.
It is not necessary here to go in detail into the relationship between longitudinal and transverse be(1) is merely the statement that only the electro. where Bim1 is the weight function for the curren vacuum fluctuations. ANDERSON which in a conductor is W 2 = U 2 — £ 2 =0)j.4/> (py^SAS. Acta 24. (longitudinal and time components). (2). it is. we must correct for the induced fields and currents. It will be convenient to quantities. (2) third component of a massive vector boson of which By wellknown arguments of gauge invariance. Schafroth. if we were confined to the plasma as we —guyp2. The only difference from an ordinary polarizable "vacuum" with bare response (5) is that in that case as p — 0 » K'^[«/(l+W)]/>2. How. then. its field A/ induces a current j„ which in turn acts as the source for an internal field A /: Thus. j = —amp^k (transverse components). If we insert a test particle. or (5). we expect ••afA. might we try to determine whether. but in a Lorentzcovariant theory of the vacuum it would be indistinguishable from the jli(k.u)Ay(k. plasma. In a truly relativistic situation such as our normal picture of a vacuum. (7) s M. the relativistic case it must be proportional to p^p. Taking into account the interaction. the new response to an applied perturbing field (9) is very like that of an ordinary polarizable medium. (S) This will turn out to be identical to Schwinger's criterion. the total field is modified to Aj=+4*jjp. (3) finite? As far as we can see. W. i „ = —aep^A. *>2>0. In the limit p—*Q both waves magnetic current can be a source of the field. This is reasonable in the problem—determine a second response function.cK)lAS. Then the response is diagonal: Kt. in the gauge (3).= —gllvK. this is not possible. Helv.154 440 P. however. (4) BL= jpSAp (9) In an insulator the response is not relativistically invariant.. (11) j^KUS+A. The longitudinal plasmon required for general gauge invariance and charge is generally thought of as entirely an attribute of the conservation according to the usual arguments. and we get •jM= K'SA„'= A'[> 2 /(/> 2 +4rf)]5. Since we cannot turn off the interactions. lead to a response A? given by 4ir •<4i = ]f = (c\ 4ir _ ~~J» up is the usual plasma frequency (AwneP/M) . If the insulator has magnetic polarizability am and electric ae.. Phys. &nd equivalent arguments give one the same are to the vacuum and could only measure renormalized 5 form in superconductivity. 645 (1951). . before consider. A had been a massless gauge field and K had been p„Ar=0. R. Let us call this response function dispersion law. we do not actually observe the responses (1). so that the coefficient of p2/4ir is less than unity. the classical case because the longitudinal plasmon disapresponse of the current to a given electromagnetic or pears at a certain cutoff energy and has a different YangMills field. of the vacuum in the elementary particle photons by the medium. 2 . (10) to describe normal polarizable behavior. therefore. K„. the transverse photons are the two transverse compomust have a certain form: Schwinger points out that in nents. for simplicity. For a nonetheless. interesting to see what the criterion is in plasma with n carriers of charge e and mass M it is terms of the actual current response function to a perturbation in the Lagrangian simply (in the limit p — 0) • > (1) K^ne^/M.'). 1/! A given distribution of current j? will.= \jt/(p+4>. while the transverse ones are considered to The dynamics of the matter system—of the plasma result from modification of the propagation of real in that case. or. Schwinger expresses the unordere A. The pole at which A propagates freely occurs at a mass (frequency) nfi=f=brK. the response equations may be written. This criterion is precisely the same as Schwinger s criterion (6) Bi(mi)dm*=l. it is propagate according to (8). only the gauge turning on the effects of electromagnetic interaction.u). This can be shown by a simp dispersion argument.havior of the plasmon.a)=K^(k. The original "bare" response function was K: j„=~Klir5All.
. In this case. u P . 648 (1960). 14. Nonetheless. 345 (1961). seems capable of directly testing the value of the vacuum polarizability Pnor to renormalization. B1(m2). if not of bare charge. it needs to be demonstrated that the necessary conservation laws can be maintained. no net true charge remains localized in the region of the dressed particle. Nambu. It is likely. W. Is this necessarily always the case? The answer is no. as elsewhere. essentially. in which the original symmetry is not manifest in the observable domain. we have not discussed nonAbelian gauge groups) for giving the gauge field mass is the degenerate vacuum type of theory.11 The theorem states. 127. Phys. This theorem was initially conjectured. Weinberg. Salam. N a m b u and G. A. Rev. all of the charge is carried "at infinity" corresponding to the fact. In addition. one presumes. The only mechanism suggested by the present work (of course. which we might interpret without V "nciples l resorting to YangMills. here. here expressed in terms of a dispersion integral. AND MASS 441 product expectation value of the current as J J (2TT) 3 The Fourier transform of the corresponding retarded Green's function is our response function: r dm2 rr$Bx{m2) K'(p)= and —{Jtp. renormalized quantities. is strictly maintained. that the way is now open for a degeneratevacuum theory of the Nambu type9 without any difficulties involving either zeromass YangMills gauge bosons or zeromass Goldstone bosons. 154 (1961). the renormalization procedure is possible for any merely polarizable "vacuum. 965 (1962). 10 Y. try to turn the problem around and see what other conclusions we can draw about possible YangMills models of strong interactions from the solidstate analogs. at least for the "undressed" matter system. 110. since the superconducting electron gas has no zeromass excitations whatever. must extend all the way to m2=0. Therefore. which is the electromagnetic field. = t We IUUOW nere. conservation of particles. GAUGE I N V A R I A N C E . Phys. not only internal test charges. follow here. The polarizability of the vacuum is not generally considered to be observable6 except in its p dependence (terms of order p* or higher in K). is infinite need not bother us. Phys. 8 7 nneiplesofof Quantum Elec Quantum Electrodynamics < w York. 9 Y. because of the solidstate analogs. What properties of the vacuum are needed for it to have the analog of a conducting response to the YangMills field? Certainly the fact that the polarizability of the "matter" system. JonaLasinio. Phys. Thirring. Thus. 1958). have its maximum possible value. tni viewpoint of W. on the other hand. •><.g*. considering the superconducting analog. Chap. aside from the short range of forces j^_thefinite mass.fl. as eisewnere. e^eZ1'2. we conclude that the plasmon is a physical example demonstrating Schwinger s contention that under some circumstances the VangMills type of vector boson need not have zero m ass. while the boson which appears as a result of the theorem of Goldstone7'8 and has zero unrenormalized mass is converted into a finitemass plasmon by interaction with the appropriate gauge field. that if the Lagrangian J. and S. that all the charge carried by a quasiparticle in a plasma is actually on the surface. Either in the case of the polarizable vacuum or of tn e "conducting" one. it is not obvious how to characterize such a case mathematically in terms of observable. I should like to close with one final remark on the Goldstone theorem.. since that is unobservable. Note that the situation does not resemble the case of "infinite" charge renormalization because the infinity in the vacuum polarizability need only occur at p2=0. Goldstone. no lowenergy experiment. Rev. 827 (1958). the ~ (Academic Press Inc. well known in the theory of metals. the coupling rather than the spectrum being affected by the screening. we can remove (11) entirely by the conventional renormalization of the field and charge Ar^AZ1'2. These two types of bosons seem capable of "canceling each other out" and leaving finite mass bosons only. In fact. 117. (aside from a factor 4ir which Schwinger has not used in his field equation) his criterion is also that the polarizability a'. 1. It is not at all clear that the way for a Sakurai3 theory is equally uncluttered. Thus. but only because we can get outside them and apply to them true electromagnetic fields. fml WmK'{p)= {pj. the fermion mass is finite because of the energy gap. J. In the normal plasma even the final spectrum extends to zero frequency. Anderson. In that case. then. The same is true of the charged Bose gas. and even possibly no highenergy one. can be shown to be precisely Z=l4ira'=ldm2 Bx(m2). Rev. Goldstone. Z.g„f) v^a [dm' J Thus. In physical conductors we can see it." but not for the special case of the conducting "plasma" type of vacuum. without taking into account the interaction with the gauge field. obviously. via the work of Nambu10 and of Anderson. 122. More serious is the implication—obviously physically from the fact that a has a pole at p2=0—that the "matter" spectrum. Rev.155 PLASMONS. jr^jZ1'2. We can. Nuovo Cimento 19.
leading to instability rather than finite . Taylor for calling my attention to Schwinger's work. thus. on the other hand. degenerate with other ground states. then. upon closer examination. and Dr. 14 L. do not in any true sense really have zero mass. is whether it is possible to describe a truly strong conservation law such as that of baryons with a gauge group and a YangMills field having finite mass. 13 possesses a continuous symmetry group under which the ground or vacuum state is not invariant. particularly. and possesses spin waves. liquid helium violates (in a certain sense only. John R. R. It is noteworthy that in most of these cases. Jeans. We conclude. that state is. John G. Spin waves also are known to interact strongly with magnetostatic forces at very long wavelengths. Soc. P. of course) gauge invariance. therefore. Klauder for valuable conversations and. 101.14 for rather more obscure and less satisfactory reasons. Phys. but a similar phenomenon happens with phonons. This implies a zeromass boson. that the Goldstone zeromass difficulty is not a serious one.W. R. when the phonon frequency is as low as the gravitational plasma frequency. Superconductivity is a familiar example.156 442 P. I should like to thank Dr. and possesses phonons. Trans.12 Utiyama13 and Feynman have pointed out that gravity is also a YangMills field. because we can probably cancel it off against an equal YangMills zeromass problem. Thus. Rev.R. H.ANDERSON mass. the solid crystal violates translational and rotational invariance. and would have a zeromass collective mode in the absence of longrange Coulomb forces. London 101. because of the peculiar sign of the gravitational interaction. superconductivity violates gauge invariance. Rev. for correcting some serious misapprehensions on my part. the Goldstone bosons do indeed become tangled up with YangMills gauge bosons and. What is not clear yet. ferromagnetism violates spin rotation symmetry. Phil. Utiyama. . Roy. Feynman (unpublished). 157 (1903). Walker. 105. and possesses a longitudinal phonon. 1597 (1956). 390 (1957). It is an amusinc observation that the three phonons plus two graviions are just enough components to make up the appropriate tensor particle which would be required for a finitemass graviton. Phys. 12 J. (4wGp)112 (wavelength ~10 4 km in normal matter) there is a phonongraviton interaction: in that case.
thereby guaranteeing invariance under both local phase and local y 5 phase transformations. We shall then examine a particular model based on chirality invariance which may have a more fundamental significance. We shall break the symmetry by fixing (cp) * 0 in the vacuum. Brout Faculte des Sciences.157 BROKEN SYMMETRY AND THE MASS OF GAUGE VECTOR MESONS* F. mt Id H M M (1) where (p = {<px +i(p2)//2. Belgium (Received 26 June 1964) It is of interest to inquire whether gauge vector mesons acquire mass through interaction 1 . despite the fact that the vector meson acquires m a s s . A calculation performed in lowest order perturbation theory indicates that those vector mesons which are coupled to currents that "rotate" the original vacuum are the ones which acquire mass [see Eq. 4 " 6 A characteristic feature of such theories is the possible existence of zeromass bosons which tend to r e s t o r e the symmetry. we shall not study the particular mechanism by which the symmetry is broken but simply assume that such a mechanism exists. This example should be considered as a rather general phenomenological model. The importance of this problem resides in the possibility that stronginteraction physics originates from massive gauge fields related to a system of conserved currents. As such. =ieA <p*d (pe2tp*wA A . (1) Lest the simplicity of the argument be shrouded in a cloud of indices. we shall show that in certain cases vector mesons do indeed acquire mass when the vacuum is degenerate with respect to a compact Lie group. Here we begin with a chiralityinvariant Lagrangian and introduce both vector and pseudovector gauge fields. We shall show in this case that the pseudovector field acquires m a s s . 7 ' 8 We shall show that it is precisely these singularities which maintain the gauge invariance of the theory. In the last paragraph we sketch a simple argument which renders these results reasonable. Theories with degenerate vacuum (broken symmetry) have been the subject of intensive study since their inception by Nambu. the phase transformation of a charged boson. Bruxelles. we first consider a oneparameter Abelian group. 3 In this note. Englert and R. by a gauge vector meson we mean a YangMills field 2 associated with the extension of a Lie group from global to local symmetry. we then present the generalization to an arbitrary compact Lie group. for example. We shall first treat the case where the original fields are a set of bosons y^ which t r a n s form as a basis for a representation of a compact Lie group. The interaction between the <p and the A „ fields is H. Universite Libre de Bruxelles. with the phase chosen for convenience such that We shall assume that the application of the 321 . (6)]. In this model the gauge fields themselves may break the y5 invariance leading to a mass for the original Fermi field. r e p r e senting.
<p2 propagator. Consider now. (b)(27r) 4 ie 2 ( ? „ ?1 / ? 2 ) = T) lim 41'(x + £)y Yr4>'{x).ABfB' 6 where A and B a r e vector and pseudovector gauge fields. a a a (3) =e W  (2) Consider the interaction Hamiltonian T r We have not yet constructed a proof in arbit r a r y order. Suppose that in the vacuum (cpg^^O for some B'.AC'{(pC') . The conventional t e r m s do not lead to a mass in this approximation if gauge invariance is carefully maintained. the similar appearance of higher order graphs leads one to surmise the general truth of the theorem. NUMBER 9 PHYSICAL REVIEW LETTERS 31 AUGUST 196i theorem of Goldstone. 1). (9) iP'{x) = exp[ifXooiiB This gives for the polarization tensor of the 322 . 5 \^c. however.B'. the same calculation as before yields n a u (q)=i(2ir)4\2({cp)TaTa(cp)) x[ Here we have used for the propagator of <p2 the value [i/{2Ti)l}/q2.(277) J 4 A. The Lagrangian is invariant under the transformation 10 int LI'5 n 'fiT n (7) ^A^a. Thus the fermion propagator S(p) is SHp) =rPX(p) with =y£[lE 2 (£ 2 )]2 1 (/> 2 ). (2) both maintains gauge invariance (n q = 0) and causes the A „ field to acquire a mass vv/*l (6) giving a value for the mass H2 = {(cp)T T(<p)). as in Johnson's model. We calculate the vacuum polarization loop n for the field A in lowest order perturbation theory about the selfconsistent vacuum.AB^B')Ta. We take into consideration only the brokensymmetry diagrams (Fig. longdashed line. (a)(27r)4ie^/il. Then the propagator of YJA B'Ta AB'fA We define the gauageinvariant current J using Johnson's method 12 : J 5 M by W (b) FIG. ((pj). the fact that the r e normalization constant is 1 is consistent with our approximation. 9 We then note that Eq. A^ propagator.A£a{x)Ta. in general. Salam.^d„£JX)' (4 > (8) where ca^ are the structure constants of a compact Lie group and Ta £j$ the antisymmetric generators of the group in the representation defined by the <pg. in the lowest order.<V1>2.b£cix)CacbAb. a set of bosonfield operators (p* (which we may always choose to be Hermitian) and the associated YangMills field Aa „. 1.C ({<p)T Ta(q>)) . 11 a brokensymmetry solution corresponding to an arbitrary mass m for the ip field fixing the scale of the problem. For a suitable choice of e and r] there exists.(2TT) J 4 q2 With A the coupling constant of the YangMills field. and Weinberg 7 is straightforward and thus that the propagator of the field <p2. Brokensymmetry diagram leading to a mass for the gauge field. wavy line. M 5 £0 (y)dyllY5]tp(x). T a.158 VOLUME 13. which is "orthogonal" to <plt has a pole at q = 0 which is not isolated. The vector field causes attraction whereas the pseudovector leads to repulsion between particle and antiparticle. Shortdashed line. One evaluates directly liv iiv 1 \i v 1 <((?£/} is. S x (/> 2 )#0 and m[lE2(m2)]Zl(m2) = 0.
Schwinger. while the usual spurious "photon m a s s " drops because of the second t e r m in (10). A. 191 (1954). Rev. JonaLasinio. S. Sakurai. The mass of the pseudovector field is roughly rfm2 as can be checked by inserting into (10) the lowest approximation for r ^ 5 consistant with the Ward identity. Johnson. Bludman and A. This effect cannot vanish in the limit q — 0. which are "rotated" vacua. reference 6. A. that is <70.S J + 2S y v uo v v 5 2 I D 2(q P )(v p )(8S JdP>)y V V A A £ troduced in the first place in order to extend the symmetry group to transformations which were different at various spacetime points.159 PHYSICAL REVIEW LETTERS VOLUME 13. reference 6.0. Let us recall that these fields were in 323 . Rev. We would like to emphasize that here the symmetry is broken through the gauge fields themselves. A. 2 C . one sees that the A propagator connects to intermediate states. Salam. Rev. Nambu. Johnson. Air Force under grant No. 227. However. Phys. Rev. NUMBER 9 31 AUGUST 1964 pseudovector field n ^5('7)=. Yang and R. 122. Rochester. Ann. 0 (12) *This work has been supported in part by the U. 380 (1960). J . 12 K. N.. Phys. Nambu and G. 'K. L. 2£1> 1 5 9 7 (1956). AFEOAR 6351 and monitored by the European Office of Aerospace Research. Goldstone. a n d S . 965 (1962). Rev. j j ! l . 4 Y. Phys. ^ 6 . 5 Y. if we break gauge invariance of the first kind and still maintain gauge invariance of the second kind this reasoning is obviously incorrect. 397 (1962). 9 A./' + k ) xs(/> + k)y„rn S(p)[dS1(p)/dp )S(p)y^dip. 1963 (unpublished).) H. Phys. in Fig. A. JJJ5. The singularity in the longitudinal r c vertex due to the brokensymmetry t e r m 2 2 ^ in the Ward identity leads to a nonvanishing gaugeinvariant n n„5(<7) in the limit q . Rev. New York. University of Rochester. Phys. 8 S . Rev.72(^)?/Tr{S(/'_l'7>r^(/'"k. 345 (1961). This is seen most clearly by writing (cp. reference 6. no dynamical manifestation of these fields should appear. (N.p + \q)=Up\q)y^y^{p ^l. One might hope that such a feature is quite general and is possibly instrumental in the realization of Sakurai's program. Klein. 7 J . ' j .Y. Mills. in this case the general feature of the phenomenological boson system survives.R s a t i s + \q). Thus one expects that when the group transformations become homogeneous in spacetime. + ( 10 ) where the vertex function ^u^>yvy^ fies the Ward identity 5 q^\^{p{q. Klein.) =([Q<p2]} where Q is the group generator. Phys. Weinberg. Phys. (11) which for low q reads Q r =q y y [ 1 . 2364 (1963). Indeed. "Broken symmetry" has been extensively discussed by various authors in the Proceedings of the Seminar on Unified Theories of Elementary Particles. Phys. U R. Letters 4. Utiyama. 1. 3 (3) We present below a simple argument which indicates why the gauge vector field need not have zero mass in the presence of broken symmetry. This means that it should cost no energy to create a YangMills quantum at q = 0 and thus the mass is zero. 3 J . Thus. 1 (1960).
where V cp = 9 <P.V ^ +^ V i F ^ . L is invariant under simultaneous gauge transformations of the first kind on cpl ± icp2 and of the second kind on A . (p2(x) = tp0. (p2 and a real vector field A interact through the Lagrangian density about the "vacuum" solution <p1(x)=Q. V"(<po2)>0. University of Edinburgh. and the metric is taken as +++. as a consequence of this coupling. f±V =3 j± B 3 B V V j± =F (3) into the form a 5^ = 0. the spinone quanta of some of the gauge fields acquire mass. we note that the righthand side of (2c) is just the linear approximation to the conserved current: It is linear in the vector potential. gauge invariance being maintained by the presence of the gradient term. The simplest theory which exhibits this behavior is a gaugeinvariant version of a model used by Goldstone 2 himself: Two real 4 scalar fields cp1. the longitudinal degrees of freedom of these particles (which would be absent if their mass were zero) go over into the Goldstone bosons when the coupling tends to zero. Higgs Tait Institute of Mathematical Physics. . the remaining six components of the scalar octet combine with the corresponding components of the gaugefield octet to describe . v 0 (4) L = \(V<p1)2i(v<P2) . H a G ^ « V V = 0. Let us suppose that V'(cp02)=0. (2b) (2c) Equation (2b) describes waves whose quanta have (bare) mass 2ip0{V"(<p02)}1/2. 6 The model of the most immediate interest is that in which the scalar fields form an octet under SU(3): Here one finds the possibility of two nonvanishing vacuum expectation values. then spontaneous breakdown of U(l) symmetry occurs. M 1 M 1 M (1) V2 \xv = V2+eVl' 3 A .160 VOLUME 13. fails if and only if the conserved currents associated with the internal group are coupled to gauge fields. the gauge field always belongs to the adjoint representation. (2a) and (2c) may be transformed. 2 that Lorentzcovariant field theories in which spontaneous breakdown of symmetry under an internal Lie group occurs contain zeromass particles.eA cp2 . one encounters a variety of possible situations corresponding to the various distinct irreducible representations to which the scalar fields may belong. In passing. Scotland (Received 31 August 1964) In a recent note 1 it was shown that the Goldstone theorem. \F^=^0{^^H'1)ecpQA}. In the absence of the gauge field coupling (e = 0) the situation is quite different: Equations (2a) and (2c) describe zeromass scalar and vector bosons. The purpose of the present note is to report that. which may be chosen to be the two K = 0. and A ^ as small quantities] governing the propagation of small oscillations 508 Equation (4) describes vector waves whose quanta have (bare) mass e<p0.m a s s excitations of a superconducting neutral Fermi gas become longitudinal plasmon modes of finite mass when the gas is charged. respectively. Edinburgh. NUMBER 16 PHYSICAL REVIEW LETTERS 19 OCTOBER 1954 BROKEN SYMMETRIES AND THE MASSES OF GAUGE BOSONS Peter W.d A v ii . 5 When one considers theoretical models in which spontaneous breakdown of symmetry under a semisimple group occurs. Eqs. Acp2. (2a) {d24cp02V"(cfio2)}(Acp2) = 0. This phenomenon is just the relativistic analog of the plasmon phenomenon to which Anderson 3 has drawn attention: that the scalar z e r o . / 3 = 0 members of the octet. Consider the equations [derived from (1) by treating A^j. by the introduction of new variables B =A • (e<P 0 ) 1 3 G IXV (A^). 7 There are two massive scalar bosons with just these quantum numbers. e is a dimensionless coupling constant.
42 (1963). It should be understood. and the 7 3 =±1 components of a 7 = 0. Rev. 1 = ^ state. J. Phys. Letters JJ3. 7 These are just the parameters which. Rev. to be published. W. was proposed for the K meson (725 MeV) by Y. 2 4 In the present note the model is discussed mainly in classical terms. 6 See. lead to the GellMannOkubo and electromagnetic mass splittings: See S. Brout. Nambu and J . T h e t w o y = 0. W. Letters U. that the conclusions which are presented concerning the masses ot particles are conjectures based on the quantization of linearized classical field equations. Rev. Glashow. Letters 13. leaving the photon as the only m a s s l e s s vector particle. 7=0 gauge fields remain m a s s l e s s : This is associated with the residual unbroken symmetry under the Abelian group generated by Y and 73. Phys. M. J . It is worth noting that an essential feature of the type of theory which has been described in this note is the prediction of incomplete multiplets of scalar and vector bosons. Goldstone. Phys. 965 (1962). Phys. Higgs. 9 ' P . Weinberg. L. e s sentially the same conclusions have been reached independently by F .Y. Anderson. the doubly charged excitation responsible for the quantization of magnetic flux is then the surviving member of a U(l) doublet. 42 (1964). 437(1961). 509 . 127. Rev. Sakurai. 134. Such a role. Ann. Rev. A detailed discussion of these questions will be presented elsewhere. J. P h y s . 321 (1964): These authors discuss the same model quantum mechanically in lowest order perturbation theory about the selfconsistent vacuum. Phys. one of these gauge fields will acquire m a s s . It may be expected that when a further mechanism ( p r e sumably related to the weak interactions) is introduced in order to break Y conservation.. Nuovo Cimento JL9. NUMBER 16 PHYSICAL REVIEW LETTERS 19 OCTOBER 1964 massive vector bosons. nothing is proved about the quantized theory. 154 (1961). 8 It is to be expected that this feature will appear also in theories in which the symmetrybreaking scalar fields a r e not elementary dynamic variables but bilinear combinations of F e r m i fields. 7=1 triplet whose mass is entirely electromagnetic. 9 In the theory of superconductivity the scalar fields are associated with fermion pairs. However. Englert and R. Phys. therefore. Brown. 5 In the theory of superconductivity such a term arises from collective excitations of the Fermi g a s . Glashow and M. Salam. (N. if the scalar octet interacts with baryons and mesons. There are two 7 =  vector doublets. A. 210. 3 P . degenerate in mass between y = ±l but with an electromagnetic m a s s splitting between 73 = ±§. and S. GellMann. More recently the possibility that the u meson (385 MeV) may be the Y=I = 0 member of an incomplete octet has been considered by L . B671 (1964). S. for example. 439 (1963). L . Coleman and S. Rev.) 15. as an isolated y = ±l. Goldstone.161 VOLUME 13. 8 Tentative proposals that incomplete SU(3) octets of scalar particles exist have been made by a number of people.
it will be seen that this has been accomplished by giving up the global conservation law usually 585 [S'VX)1'EAV*U)' and if it is possible consistently to take Zj^ijk x (OIA^ I 0)* 0. of course. represents a departure from the a s sumptions of the theorem. briefly stated. Goldstone's remarkable theorem 1 has played an important role. then AAx) has a z e r o .m a s s particle. England (Received 12 October 1964) In all of the fairly numerous attempts to date to formulate a consistent field theory possessing a broken symmetry. It has more recently been observed that the assumed Lorentz invariance essential to the proof2 may allow one the hope of avoiding such massless particles through the in . Imperial College. a s s e r t s that if there exists a conserved operator Qz. and a limitation on its applicability which in no way reflects on the general validity of the proof. While this result might suggest a general procedure for the elimination of unwanted massless bosons. W. B. within the framework of a simple soluble field theory.162 GLOBAL CONSERVATION LAWS AND MASSLESS PARTICLES* G. Kibble Department of Physics.t. London.t C. In this note we shall show. Hagen.m a s s particle in its spectrum. This theorem. that it is possible consistently to break a symmetry (in the sense that Tjkhjk^^k 10)*°) without requiring that A{x) excite a z e r o . R.such that troduction of vector gauge fields and the consequent breakdown of manifest covariance. S. 3 This.and T. Guralnik.
and the operator algebra possesses the full symmetry. a theory which 586 ( . since superconductivity appears to display a similar behavior. However.4. (The quantization of A^ leads to an indefinite metric for one component of <p. The same effect is not present in the Lorentz gauge where zeromass excitations which preserve charge conservation are found to occur. a r e essentially those of the BroutEnglert model. only if the contributions from spatial infinity vanish. however. 6 Our starting point is the ordinary electrodynamics of massless spinzero particles characterized by the Lagrangian £ ?Vv.8 2 + 7 ? 1 2 ) A . as our example. If. <p = . causality is a requirement which must be imposed with caution. and find (8 2 + i J 1 > 1 = 0.. r = 0> 1 R where the superscript T denotes the transverse T part. and seek solutions of our model consistent only with the differential conservation laws. The two degrees of freedom of A^ combine with <px to form the three components of a massive vector field.°(x).3 A . it will not necessarily generate local gauge transformations upon^4.. However. the theory is not manifestly covariant (e. suffers from the fact that the usual canonical quantization is inconsistent with the field equations..163 VOLUME 13. This. M 0t). F^ =d^A .</=0. (These correspond to gauge parts rather than physical particles. allow the possibility of the breakdown of such global conservation laws. in such a way as to preclude the possibility of applying the Goldstone theorem. j J . and q is the Pauli matrix u 2 .yl. The brokensymmetry condition ieQq{Q\ip\Q) = r ) ^ l \ will be imposed by approximating ie§(pVq<pAu in the Lagrangian by (p^qA^. It is clear that such a modification of the basic operator relations is a far more drastic step than that taken in the usual brokensymmetry theories in which a degenerate vacuum is the sole symmetrybreaking agent. 5 and bears some resemblance to the classical theory of Higgs. Since Qj consequently may not be time independent. We consider. the relation 3 M <0l[j.U')]l0> = 0 implies that rd 3 x(OI[. it is easily seen that it is completely decoupled from the other (massive) excitations.0 i J i 3 was partially solved by Englert and Brout.) Since we choose to view the theory as being imbedded as a linear approximation in the full theory of electrodynamics. is asserted to be a consequence of the local conservation law a „ j ^ = 0. these equations will have significance only in the radiation gauge. The resulting equations of motion. of course.(x') for x°*x'° despite the existence of the differential conservation laws The phenomenon described here has previously been observed by Zumino 4 in the radiationgauge formulation of twodimensional electrodynamics where the usual electric charge cannot be conserved. the time independence of Q.. Normally.g.) We shall. NUMBER 20 PHYSICAL REVIEW LETTERS 16 NOVEMBER i9&J implied by invariance under a local gauge group. While one sees by inspection that there is a massless particle in the theory. radiationgauge electrodynamics). however. With no loss of generality.°(x. = lF^"(d A 3 A ? ^ i )+\F^F + (p 9 <p+k<p <p +ien<p qwA . we can take TJ2 = 0. = fd x. where <p is a twocomponent Hermitian field.U')]IO> = const j af =cp*ri. the possibility of breaking such global conservation laws must not be lightly discarded. The Lorentzgauge formulation. and can be solved in either the radiation 7 or Lorentz gauge. however. The consequent time dependence of the generators Qj destroys the usual global operator rules of quantum field theory (while leaving the local algebra unchanged).3 tpi)A . . is always the case in a fully causal theory whose commutators vanish outside the light cone.
164
ME 13, NUMBER 20 PHYSICAL REVIEW LETTERS 16 NOVEMBER 1964
and has nothing to do with the Goldstone t h e o r e m . It is now s t r a i g h t f o r w a r d to d e m o n s t r a t e the f a i l u r e of the c o n s e r v a t i o n law of e l e c t r i c c h a r g e . If t h e r e e x i s t s a c o n s e r v e d c h a r g e Q, then the r e l a t i o n e x p r e s s i n g Q a s the g e n e r a t o r of r o t a tions in c h a r g e s p a c e i s [Q,(p(x)] = e0q(p{x).
Our b r o k e n s y m m e t r y r e q u i r e m e n t i s then <0[Q, < P i U)]IO> = iTj o r , in t e r m s of the s o l u b l e m o d e l c o n s i d e r e d here, \d3X • 7JX<0 I K ° U ' ) , <PiU)] I 0) = IT],. F r o m the r e s u l t (0\<plo(x')<pi(x)\0) = aBA(+Hx'x;11l2),
In s u m m a r y then, we have e s t a b l i s h e d that it m a y be p o s s i b l e c o n s i s t e n t l y to b r e a k a s y m m e t r y by r e q u i r i n g that the v a c u u m e x p e c t a t i o n value of a field o p e r a t o r be nonvanishing without g e n e r a t i n g z e r o  m a s s p a r t i c l e s . If the t h e o r y l a c k s m a n i f e s t c o v a r i a n c e it m a y happen that what should be the g e n e r a t o r s of the t h e o r y fail to be t i m e  i n d e p e n d e n t , d e s p i t e the e x i s t e n c e of a l o c a l c o n s e r v a t i o n law. T h u s the a b s e n c e of m a s s l e s s b o s o n s is a c o n s e q u e n c e of the i n a p p l i c a b i l i t y of G o l d s t o n e ' s t h e o r e m r a t h e r than a c o n t r a d i c t i o n of it. P r e l i m i n a r y i n v e s t i g a t i o n s i n d i c a t e that s u p e r c o n d u c t i v i t y d i s p l a y s an a n a l ogous b e h a v i o r . T h e f i r s t n a m e d a u t h o r w i s h e s to thank D r . W. G i l b e r t for an enlightening c o n v e r s a t i o n , and two of us (G.S.G. and C.R.H.) thank P r o f e s s o r A. S a l a m for h i s h o s p i t a l i t y . *The r e s e a r c h reported in this document has been sponsored in whole, or in part, by the Air Force Office of Scientific Research under Grant No. AF EOAR 6446 through the European Office of Aerospace Research (OAR), U. S. Air F o r c e . ^National Science Foundation Postdoctoral Fellow. •"•On leave of absence from the University of Rochester, Rochester, New York. ' j . Goldstone, Nuovo Cimento W, 154 (1961); J . Goldstone, A. Salam, a n d S . Weinberg, Phys. Rev. _127, 965 (1962); S. A. Bludman and A. Klein, Phys. Rev. JJ51, 2364 (1963). 2 W. Gilbert, Phys. Rev. Letters L2, 713 (1964). 3 P . W. Higgs, Phys. Letters _12, 132 (1964). 4 B . Zumino, Phys. Letters _10, 224 (1964). 5 F . E n g l e r t a n d R . Brout, P h y s . Rev. Letters 13, 321 (1964). 6 P . W. Higgs, to be published. This is an extension of a model considered in more detail in another context by D. G. Boulware and W. Gilbert, Phys. Rev. 126, 1563 (1962).
one i s led to the c o n s i s t e n c y condition T)1exp[iril(x0'x0)} = Til,
which i s c l e a r l y i n c o m p a t i b l e with a n o n t r i v i a l T}1. Thus we have a d i r e c t d e m o n s t r a t i o n of the failure of Q to p e r f o r m i t s u s u a l function a s a c o n s e r v e d g e n e r a t o r of r o t a t i o n s in c h a r g e s p a c e . It is well to m e n t i o n h e r e that this r e s u l t not only does not c o n t r a d i c t , but i s a c t u a l l y r e q u i r e d by, the field e q u a t i o n s , which imply (3 0 2
+
r )l 2 )Q = 0.
It is a l s o r e m a r k a b l e that if Au i s given any bare m a s s , the entire theory becomes m a n i festly c o v a r i a n t , and Q is c o n s e q u e n t l y cons e r v e d . G o l d s t o n e ' s t h e o r e m can t h e r e f o r e a s s e r t the e x i s t e n c e of a m a s s l e s s p a r t i c l e . One indeed finds that in that c a s e cpl h a s only z e r o mass excitations.
587
165
PHYSICAL
REVIEW
VOLUME
155,
N U M B E R
S
25
MARCH
1967
Symmetry Breaking in NonAbelian Gauge Theories*
T. W. B. KIBBLE
Department of Physics, Imperial College, London, England (Received 24 October 1966) According to the Goldstone theorem, any manifestly covariant brokensymmetry theory must exhibit massless particles. However, it is known from previous work that such particles need not appear in a relativistic theory such as radiationgauge electrodynamics, which lacks manifest covariance. Higgs has shown how the massless Goldstone particles may be eliminated from a theory with broken U(\) symmetry by coupling in the electromagnetic field. The primary purpose of this paper is to discuss the analogous problem for the case of broken nonAbelian gauge symmetries. In particular, a model is exhibited which shows how the number of massless particles in a theory of this type is determined, and the possibility of having a broken nonAbelian gauge symmetry with no massless particles whatever is established. A secondary purpose is to investigate the relationship between the radiationgauge and Lorentzgauge formalisms. The Abeliangauge case is reexamined, in order to show that, contrary to some previous assertions, the Lorentzgauge formalism, properly handled, is perfectly consistent, and leads to physical conclusions identical with those reached using the radiation gauge.
I. INTRODUCTION
T
HEORIES with spontaneous symmetry breaking (in which the Hamiltonian but not the ground state is symmetric) have played an important role in our understanding of nonrelativistic phenomena like superconductivity and ferromagnetism. Many authors, beginning with Nambu, 1 have discussed the possibility that some at least of the observed approximate symmetries of relativistic particle physics might be interpreted in a similar way. The most serious obstacle has been the appearance in such theories of unwanted massless particles, as predicted by the Goldstone theorem.2 In nonrelativistic theories such as the BCS model, the corresponding zeroenergygap excitation modes may be eliminated by the introduction of longrange forces. The first indication of a similar effect in relativistic theories was provided by the work of Anderson,3 who showed that the introduction of a longrange field, like the electromagnetic field, might serve to eliminate massless particles from the theory. More recently, Higgs4 has exhibited a model which shows explicitly how the massless Goldstone bosons are eliminated by coupling the current associated with the broken symmetry to a gauge field. The reasons for the breakdown of the Goldstone theorem in this case have been analyzed by Guralnik, Hagen, and Kibble.5 The situation is identical with that in the nonrelativistic domain.
* The research reported in this document has been sponsored in part by the U. S. Air Force Office of Scientific Research OAR through the European Office Aerospace Research, U. S. Air Force. 1 Y. Nambu, Phys. Rev. Letters 4, 380 (1960). Y. Nambu and G. JonaLasinio, Phys. Rev. 122, 345 (1961); M. Baker and S. L. Glashow, ibid. 128, 2462 (1962); S. L. Glashow, ibid. 130, 2132 (1962). • J. Goldstone, Nuovo Cimento 19, 154 (1961); J. Goldstone, A. Salam, and S. Weinberg, Phys. Rev. 127, 965 (1962). 3 P. W. Anderson, Phys. Rev. 130, 439 (1963). 1 P. \V. Higgs, Phys. Letters 12, 132 (1964). 6 G. S. Guralnik, C. R. Hagen, and T. W. B. Kibble, Phys. Rev. Letters 13, 585 (1964). See also T. VV. B. Kibble, in Proceedings of_ the Oxford International Conference on Elementary Particles, 1965 (Rutherford High Energy Laboratory, Harwell, England, 1966), p. 19.
In either case the theorem is inapplicable in the presence of longrange forces, essentially because the continuity equation dlxj>>=Q no longer implies the time independence of expressions like fd3x [j°(x),<K0)], since the relevant surface integrals do not vanish in the limit of infinite volume. (In the relativistic case, the theorem does apply if we use the Lorentz gauge; but then it tells us nothing about whether the massless particles are physical.) It should be noted that the extension or corollary of the Goldstone theorem discussed by Streater6 also fails in precisely this case. If longrange fields are introduced, spontaneous symmetry breaking can lead to mass splitting. As has been emphasized recently by Higgs,7 it thus appears that the only way of reconciling spontaneous symmetry breaking in relativistic theories with the absence of massless particles is to couple in gauge fields. Another possibility is that Goldstone bosons may turn out to be completely uncoupled and therefore physically irrelevant. In this case, however, the Hilbert space decomposes into the direct product of a physical Hilbert space and a freeparticle Fock space for the Goldstone bosons. The broken symmetry appears only in the latter, and no trace of it remains in any physical quantities. In most simple cases, the symmetry transformations leave the physical Hilbert space completely invariant; and in any case they act unitarily on it. Such theories cannot therefore explain observed approximate symmetries. This decoupling of Goldstone modes does occur in the Lorentzgauge treatment of models like that discussed by Higgs, in which in fact no trace of the original £/(l) symmetry remains in the physical states. However it does not occur in corresponding nonAbelian gauge theories, to which the conventional (i.e., GuptaBleuler) Lorentzgauge formation is inapplicable. It has been suggested by Fuchs 8 that in the case of nonAbelian gauges the massless particles may persist
6 7 8
R. F. Streater, Phys. Rev. Letters 15, 475 (1965). P. W. Higgs, Phys. Rev. 145, 1156 (1966). N. Fuchs, Phys. Rev. 140, B911, (1965).
155
1S54
166
155
SYMMETRY
BREAKING
IN
NONABELIAN
GAUGE
THEORIES
1555
even after the introduction of gauge fields. His argument is based on the use of the Lorentz gauge, and Schwinger's extendedoperator formalism.9 His conclusions disagree with those reached by Higgs4, using the radiation gauge. However, Fuchs has already remarked that his method leads to considerable difficulties of interpretation inasmuch as the energy spectrum is not bounded below. This is not the only difficulty which the method encounters. In order to bring out clearly the relationships between the Lorentzgauge and radiationgauge treatments, we shall begin by reexamining, in Sees. II and III, the simple Abeliangauge case. The Coulombgauge treatment given in Sec. II contains a summary of some of Higgs's results, reexpressed in a form appropriate to the comparison with the Lorentzgauge treatment given in Sec. III. In particular, we aim to show that the correct treatment of U(V) symmetry breaking in Schwinger's extendedoperator formalism does not involve any alteration in the "subsidiary condition" which selects the gaugeinvariant physical states. (This condition is changed in the method proposed by Fuchs.) In Sec. IV we go on to discuss the generalization of the model treated by Higgs to an arbitrary nonAbelian group. Our aim here is not to give a complete discussion of this model but mainly to show how the number of massless fields in the theory is determined, in terms of the "canonical number" introduced by Bludman and Klein.10 Finally, in Sec. V, we exhibit a model involving spontaneously broken U(2) symmetry which is entirely free of massless particles, and moreover, in which the physical states retain clear indications of the underlying symmetry. As in all such theories, the most obvious indication of symmetry breaking is the appearance of an incomplete multiplet of massive scalar particles. Our results lead to the following conclusion: If all the currents associated with a broken nonAbelian symmetry group are coupled to gaugevector fields, the number of massless vector bosons remaining in the theory is just the dimensionality of the subgroup of unbroken symmetry transformations. In particular, if there are no unbroken components of the symmetry group, then no massless particles remain. II. COULOMB GAUGE It will be useful to begin by summarizing in rather different language some of the results obtained by Higgs.7 We start with the Goldstone model: a complex scalar field <f> described by the Lagrangian L = 0**a M 0+0*d M **0^ M F(0*0). (1)
Consequently, the current
i"=iefo"*0W*)
satisfies 3„i"=0.
(3)
(4)
If V{(f>*4>) has a maximum at < * £ = 0 and a minimum £< elsewhere, then we may expect that the expectation value of 0 in the vacuum (or ground state) is nonzero. From the equations of motion we obtain the consistency condition (dV/d4>*)=(<t>V'(<t>*4>)) = 0, (5) which serves to determine the magnitude of (<j>). If {4>) = r\ is a possible solution, then so is {<f,) = neia. There is, therefore, an infinitely degenerate set of vacuum states, parametrized by the phase a. Formally, transformations from one to another are generated by the "unitary operator" eiXQ with
Q=jd>xf(x).
(6)
However, when (</>)s^0 the integral here is divergent, and the various degenerate vacuums belong to unitariiy inequivalent representations. The Goldstone theorem requires the existence of massless particles in this theory. They may be exhibited by making the polar decomposition *=21'2pe*'*, (7)
introduced by Higgs. (We shall ignore problems of operator ordering.) The canonically conjugate variables are the time components of the vectors
p*= 2 1 ' 2 (^*e«+0Mg.'J) )
and 8" = 2"1 i"ip (0*V *  <t>"e^) =  elj".
(8)
It should be noted that p, p", and #" are all invariant under the transformations (2) while # transforms according to d(x)+d(x)+e\(x). (9) This shows, incidentally, that there is no fundamental distinction between transformations expressible as rotations or translations of the field variables, since one may be converted into the other by a change of variables. In terms of the new variables, the Lagrangian becomes L = (fd„P+d»dl$%p<'PtlWl,/2fF(ip 2 ). (10)
The brokensymmetry condition is expressed by setting P=M+P', (P')=O. (ii)
This is clearly invariant under the constant gauge transformations 0(x)>e"'**(*). (2)
9 10
J. Schwinger, Phys. Rev. 125, 1043 (1962); 130, 402 (1963). S. A. Bludman and A. Klein, Phys. Rev. 131, 2364 (1963).
Clearly, the p field describes particles whose mass is determined (to lowest order) by the second derivative of V at P =  T ;  , while the # field describes massless particles.
167 1556 T. W. B. K I B B L E 155
Now let us consider the coupling of the current (3) to the electromagnetic field. We have11 L=F'»dyAll+iF<"Fll, +(j>"*(d„+ieAfl)<j)+<^(dliieAll)<i>* V*4>»V(.4>*<t>),
tional invariance.) It may be noted that the equation obtained by variation of &, namely V 2 C=0, (20)
(12)
which is invariant not only under (2) but also under the gauge transformations <t>(x) > e"x(l>#(a:), A,, (x) —* A^{x) — c^X (x). (13)
shows similarly that C is at most a constant. From the structure of (17) we see that no massless particles remain in the theory. The mass of the scalar particles described by p is as before determined to lowest order by the second derivative of V at />= ij. Now, however, the massless particles described by 0 have become the longitudinal modes of the vector field described by Bf, whose mass is to a first approximation It may be worth recalling the reasons for the failure of the Goldstone theorem to apply in this case. The essential point is that the continuity equation (4) no longer implies the time independence of the commutator
In terms of the polar variables introduced in (7) and (8), this Lagrangian becomes L=Fi»d,All.+\F>"Ft.+i?dlj>+Wdl&ip>'Plt WjlfVQfi+eAp. (14)
We now wish to investigate the relationship between the Coulombgauge and Lorentzgauge treatments of this Lagrangian. Let us consider first the Coulomb gauge. To preserve the analogy with our later treatment of the Lorentz gauge, it will be convenient to impose the Coulombgauge condition by adding t o l a Lagrange multiplier term CdkAK (15) This destroys the invariance under (13), but not that under the constant gauge transformations (2). It is now convenient to introduce new variables B^A,+eld^. Then, using the equation d^—epB,, to eliminate t?,, from the Lagrangian,
t i 12
J <Px<£?(x)M0)l)
(21)
(16)
because the relevant surface integral fails to vanish in the limit of infinite volume. Although the operator (6) does not exist, its commutators do exist in a formal sense provided we perform the space integration after the evaluation of the commutator. In the absence of longrange fields Q is timeindependent (in the sense that these commutators are so), but, as was pointed out by Guralnik, Hagen, and Kibble,5 this is no longer true when gauge fields are present. This is easy to verify for our particular model. We have f=eVB", whence <[i*to,<?(0)])= ^{lMd„(SyB(x),BKO)l)(23) (22)
we obtain (17)
L=F>»d,Bl.+lFi"Flu,+ie p BllBi'+P>'di,p ip"pltV(^)Cdk(Bkeidkd).
This Lagrangian is still invariant under (2), or its equivalent d(x)+d(x)+e\, (18) but in a completely trivial way. In fact, & is determined by W}=edkBk, (19) and (18) represents merely the arbitrariness in the solution of this equation. (Explicit dependence on the coordinates is ruled out by the requirement of transla11 This model, or a closely related one, has been discussed in Refs. 4, 5, and 7 and also by F. Englert and R. Brout, Phys. Rev. Letters, 13, 321 (1964). 12 This is permissible since it is an algebraic equation fori?,. One is not allowed to solve an equation of motion and substitute the results in the Lagrangian, but one is allowed to substitute explicit solutions obtained without integration. See, for example, T. W. B. Kibble and J. C. Polkinghorne, Nuovo Cimento 8, 74 (1958).
This form clearly exhibits the nonlocal structure of ([_/',#]). Inserting a Lehmann spectral form on the righthand side of this equation, it is easy to see that {Lj°>&^}) i s causal but that its space integral is not timeindependent. Indeed in lowest order it is — e cos(e\ri\x°). i n . LORENTZ GAUGE Now let us turn to the description of this model in terms of the Lorentz gauge. Following Schwinger,9 we may impose the Lorentzgauge condition by adding to L a Lagrange multiplier term Gd^+iaGG, (24)
where a is an arbitrary constant introduced to allow direct comparison both with Schwinger's formalism (a=0) and with the more conventional formalism (a= 1) adopted by Fuchs. Note that in secondorder form (24) is equivalent to iflr'CM")', (25)
so that a= 1 corresponds to the usual Fermi Lagrangian.
168 155 SYMMETRY BREAKING IN NO] J  A B E L I A N
GAUGE THEORIES 1557
The advantage of the firstorder form (24) lies precisely in the possibility of taking a = 0 . The generator of the gauge transformations (13) is G(X) = d*x[\(x)d°G(x)G(x)d°\(x)']
To exhibit the covariance of the Lagrangian it is convenient to introduce also new dependent field variables Bo, G', and i//', so that we may write L= F»"d„Bll+lF>"Ftl„+im,2B>'B)1 +GdllG+i"d„^i^ll+mi'Gll+iaGG. (31)
=
<Px[\(fdkF°k)Gd<>\].
In Schwinger's extendedoperator formalism the physical states are distinguished by the gaugeinvariance requirement G(X)¥=0, (26) or equivalently, G*=0, (FdkFo^V^O. (27)
On the other hand, in the more familiar GuptaBleuler formalism,13 only the positive frequency components of G are required to annihilate the physical states. Both these formalisms will be considered in the sequel. *=(f,G°',S/>. (32) The important difference between the Coulomb and Lorentz gauges lies of course in the number of degrees The canonically conjugate variables \p°, G, and F0' are of freedom. Since (15) involves no time derivatives, C represented as functional derivatives with respect to the is not a dynamical variable. However, in the Lorentz appropriate variables. Then the functionals representing gauge, the canonically conjugate pairs of variables are physical states are distinguished by the gaugeinC4,,F°*), (G,A°), (p,p<>), and (t?,t?°). The pfield excita variance requirement (26) or (27), which may now be tions are essentially irrelevant to our discussion. So for written simplicity we shall make the approximation of replacing GV=i(6/5G°')y=0, p by 11\ . This should be a good first approximation if (33) the mass determined by the second derivative of V is large. Thus we have to consider the Lagrangian It follows that * is actually independent of the two scalar fields, and may be represented by a functional of L = F>"d,AIL+iF>"FIL<,+d»dltd Wll/2\v\*+eAlli}>'~GdrA>+iaGG. (28) Bi alone. This is, of course, in conformity with the conclusion reached in the Coulombgauge treatment, Since this Lagrangian is only quadratic in the field that only vector particles appear in the physical states. It should be remarked that the symmetry breaking variables, the resulting theory is exactly soluble. As before, it is convenient to introduce new variables. corresponds to having a nonzero value of i\ and therefore of the vector particle mass m, and has nothing whatWe write w 2 =e s h 2 (29) soever to do with the conditions (33) which should be imposed whether or not the symmetry is broken. The and make a canonical transformation to the pairs physical states are still gauge invariant in the sense oi (Bi,F ), (G,G°), and (M°), where described by (33) even when the original symmetry (2) is broken. Indeed, in this formalism, the localgauge 1 1 transformations (13) do not act on the physical states £,=4,T3,IH d,G, at all, although of course the global transformations (2) e w2 do. This is associated with the wellknown fact that 1 while the oneparameter group of gauge transformations 0 k G<>=A +—dkF> , (30) (2) yields a conservation law for the electric charge, the infiniteparameter group of local transformations (13) does not yield an infinite set of physical conservation laws. 1 1 The symmetrybreaking discussed by Fuchs8 is the k and yffi=—!?°+—dkF> . breaking of the conditions (33). This means that more hi m states than usual are admitted as physical. This 13 See, for example, J. M. Jauch and F. Rohrlich, The Theory of procedure has a number of grave disadvantages, notably Photons and Electrons (AddisonWesley Publishing Company, the fact that when (33) is relaxed, the energy spectrum Inc., Reading, Massachusetts, 1955), Chap. 2.
Clearly, we have a vector field describing particles of mass m, and two scalar fields (in addition to field p corresponding to the suppressed modes). The vector particles are, of course, precisely those found earlier in the Coulomb gauge. To discuss the physical significance of the scalar fields in (31), we have to be more precise about the conditions on physical states. Let us first consider Schwinger's extendedoperator formalism. Here the fields are not to be regarded as operators in a Hilbert space; they are extended operators acting on a more general space of functionals in which no scalar product is defined. The states are labeled by the eigenvalues of, for example, $, G°, and Bi, all at one time h:
169 1558
T. \Y
KIBBLE
155
is no longer bounded below. It should also be noted that (33) provides the only guarantee that the equations of motion agree, for physical states, with those obtained from the original gaugeinvariant Lagrangian. To relax this condition is not merely to break a symmetry but to change the physical equations of motion. It may be that symmetry breaking of this type has some physical relevance, despite these difficulties. What we are concerned to show here, however, is that it is not the method which is the true Lorentzgauge analog of the Coulombgauge formalism described in the preceding section. To achieve agreement between the two approaches it is necessary (and sufficient) to break the symmetry under the transformations (2) while maintaining the conditions (33) intact. It is interesting to examine in more detail the unphysical fields G and <p, which satisfy the field equations 9^G= —w^v, duG»=aG, whence DG=0, D^= amG.
(35) (34)
replace e(k°) on the right by 6{k"). However, in the neighborhood of k=0 this is not correct. For, although (d/dK%e(k°)2Tr&(k,2K2)2 yields a welldefined distribution in the limit K2 — 0, the corresponding symmetric > expression (d/3/<2)[2rf(&2—K2)] does not.14 In fact, integrating it over a small volume around k = 0, one finds a result which diverges like ln«2 as K2 — 0. To > obtain a welldefined distribution, one must subtract off this infinite term, and consider for example the expression —[2T5(/fe2/c2)]27r2 In—5 4 (£) , dK2 M2 for any constant mass M, which does possess a welldefined limit as K2 — 0. Thus, we obtain > £p„<«\K0)0)2(27r)^„*) = 0(£o)27ra(£2)o:m:
,3K 2
{0(£°)2T6(fc2/c2)}
From these equations and the canonical commutation rules, we can easily derive covariant commutation relations. The consistency of the conditions (33) for different times is assured by the relation [G(*),G(0)>0. We also find [_G(x),i(0)2=~imD(x) and finally, THx)$ (0)1 = iD(x)iam2\ — A(X,K2) Ld/c2 J, 2 _o = ~ e(«°)8(a?),
2TT
7r 2 ln M2
8i{k)
L
(40)
(36) (37)
=
2x
e M p ^ J + i r f M ] . (38)
In the extendedoperator formalism, these relations are not required to possess a representation in a Hilbert space, and pose no particular problem. However, let us now examine the formalism of Gupta and Bleuler, in which the fields are operators in a Hilbert space of indefinite metric, and the physical states are selected by the condition G<+'> = 0. (39)
Then, taking the vacuum expectation value of the commutator (38), and denoting the diagonal elements (±1) of the metric operator by pn, we find £ Pn (n\f{0)\0)\*(2*)<£6t(j,nk)6<(Pn+k)l = t(k<,)2Tr5{ki)airi
 d
&K2
e(#027ra(£2»<2)}
For k?^0, we can drop the second term in the square brackets on the lefthand side of this equation, and
The arbitrariness in the constant M arises from the fact that division of a distribution equation by e(&°) yields a result arbitrary to the extent of a multiple of St(k). This arbitrary constant may also be regarded as a manifestation of the arbitrary additive constant in the field <//, since a field translation of i/' would clearly change the lefthand side of (40) by a multiple of 84(&). It is evident that unless a = 0, the matrix elements of tp must be extremely singular. The basic reason for this may be seen from the equations of motion (35). Since the field G has a singularity at k2=0, like 1/k2, we see that \f/ must have an even worse singularity, like (1/&2)2. The structure in the right side of (40) is itself well defined14 (for given M) but can only be obtained by a cancellation between infinite positive and negative terms on the left. It is in this sense that the conventional Lorentzgauge treatment (which implies a = l ) is inconsistent. It is very interesting that the particular case a = 0 does not encounter these problems. This case is interesting from another point of view also. Since this theory is manifestly covariant, the Goldstone theorem is certainly applicable. In general, the massless bosons it predicts are described by the field G [since dfi is the conserved current; note that the proof of the theorem rests on the commutator function (37)]. However, in the special case a = 0 , there are two independent massless fields corresponding to the fact that in that case the Lagrangian is invariant (up to a divergence) not only under the transformations (2) but also under G(x) — > G(*)+X.
14
J. Girding and S. Lions, Nuovo Cimento Suppl. 14, 9 (1959).
170 155 SYMMETRY BREAKING IN NONABELIAN GAUGE THEORIES 1559
We note that Schwinger's extendedoperator formalism is more flexible, even in the simple Abeliangauge case, than that of Gupta and Bleuler, which works properly only for a = 0. In the case of nonAbelian gauges, the GuptaBleuler formalism is wholly inapplicable, because the analog to G no longer satisfies the free wave equation, and cannot therefore be resolved covariantly into positive and negative frequency components, so that (39) becomes meaningless. IV. NONABELIAN GAUGE MODELS Let us now consider a model of an wcomponent real scalar field fa which transforms according to a given wdimensional representation of a compact Lie group G of dimension g, and (41)
However, not all these will be independent in general, since there may be a subgroup G, of G which leave ij invariant (the isotropy group of G at ij 1! ). This is the subgroup corresponding to symmetries which are not broken. Let v—the "canonical number" of Bludman and Klein10—be the number of algebraically independent invariants constructible from r\, or equivalently the number of algebraically independent invariant "Hartree conditions" (46). We assume that q can be brought to a canonical form in which only v of its components are nonzero. Further, we assume that none of these components is accidentally zero. Then the set of equivalent solutions (fa) forms a manifold of dimension r = n— v. It is clear that this manifold may be identified with the factor space G/G, (not in general a group). In fact, we may write the representative of each element of G in the form eeV
T
\T=\ATA.
(47)
Here the XA are g real parameters and the TA are g real antisymmetric nXn matrices obeying the commutation relations of the associated Lie algebra [TA,TB] = TctcAB. (42)
where e" is an element of the (g—r)dimensional subgroup G„ and the remaining r parameters p. serve to parametrize the solutions (<j>) by. the identification (fa = e"Tv. (48)
These relations are satisfied in particular by the matrices lA = {t°AB) of the adjoint representation. The Lagrangian density is taken to be L=fad„<f>kfafaV(fa, (43)
Since these solutions are physically equivalent (though, of course, unitarily inequivalent) there is no essential loss of generality in choosing <*> = *. (49)
where fa transforms like fa and V{<j>) is invariant, under It will be convenient to adopt a set of coordinates in (41). (The notation implies the use of an invariant which the first r elements of i\ are zero, while the last v scalar product in the ^dimensional space of the vari elements are not. It is then useful to make a correspondables <j>.) From the invariance of the Lagrangian we may ing "polar decomposition" of <j>, analogous to (7). We infer the existence of conserved currents, write (50) <l> = e»rp = e»'T(r,+p'), JA»=<t>»TAfa (44) In any finite spacetime volume, the transformation where 0, like p, has r components, while the first r components of p are zero. We shall distinguish these (41) is generated by the operators components by using indices a, b, ••• = !, • • •, r and a, /3, • • • = r + l , •••, n. (45) \A<PXU°(X). Consider the action of the generators TA on p. We note that TAp = 0 for those indices A belonging to elements of the Lie algebra of G,. A nonzero result However if <j> has a nonvanishing expectation value so occurs only for T^, <z=l, • • • ,r. Moreover, consider the that the symmetry is broken, then as usual the integrals matrix over all space do not exist, and the transformation (41) Xa"={TaPy, (51) is not unitarily implementable. The expectation value (fa in a translationally in which consists of the components of these vectors in the variant (vacuum) state is restricted by the consistency subspace in which p itself vanishes. We assert that this condition matrix is nonsingular. For, if not, we can find some (dV/d(t>) = 0. (46) linear combination of the generators Ta which gives zero acting on all vectors p. But then this linear comIf (<£) = n is a consistent brokensymmetry solution then bination should be an element of the Lie algebra of G„ KT so also is {4>) = e ' i), for any X. If we choose a particular which it is not. ri, then all other physically equivalent solutions may be 16 expressed in this form. [There may, of course, be other See, for example, R. Hermann, Lie Groups for Physicists physically inequivalent disjoint solutions of (46)]. (W. A. Benjamin, Inc., New York, 1966), p. 3.
171 1560 T. W. B. N I B B L E
155
The canonically conjugate variables to d and p are, as in (8), the time components of the 4vectors #a» = 4»<e»TTbpk\, p ^ =(*»«* ' r ) „ , where A».= [ ( l  « f  * 1 ' ) / *  ' ] ' . . This follows from the relation /le^'N6 eXi=rB • . (55) e xr d\A \ \i IA Note, however, that we have denned A as an rXr submatrix of the gXg matrix appearing in (55). This is permissible because, as we have seen, TBp = 0 unless B is one of the first r indices. Thus, in place of TBpABa, we can write TbpA.ha. The currents (44) may all be expressed in terms of the canonical variables t? and iK In fact d j^=<V(A 1 )° i ,(<r" < )V (56) (54) (52) (53)
analogous to (15). It is now convenient to introduce new variables B\= and GAr=FBr{f?t)BA. Then, elimination of the variables d* yields the Lagrangian L=iGA^dyB^d»BA,lABCBBltBcl) + i a 4 " ' G V + p " < v   p » p , 1  v(P) +P»TapB\+UX°bB\)2. {e« •') ABA V h [ (1  e° • <)/# • f\Ahd$b (61)
(62)
It is clear from this form that g—r of the g vector fields have zero mass and do not interact directly with the fields p, while the remaining r have masses given in lowest order by the mass matrix Mab=(vTa)c(Tbr,y, (63)
In terms of the new variables the Lagrangian (43) may be written L=p^,P\p^p,V{p) + ! ? 0 *^i[{*V(A 1 )° 6 p*:r 6 p} (X 1 ) V] 2 . (57)
Here we have used explicitly the fact that X, as defined by (51), is nonsingular. This Lagrangian, of course, retains the in variance under (41). Clearly, p and p" are invariant. The effect on# of an infinitesimal transformation may be written as 5 1 )'=(A 1 )« i (e' 1 ) , J^, (58)
As an illustration of the discussion in the preceding section, we shall consider here a simple model of broken The masses of the particles described by the fields p' 17(2) symmetry in which no massless particles remain. would be principally determined, if the interaction is The model contains a complex threecomponent field weak, by the second derivatives of V near the point 0 = (<pi) and four vector fields A,, and A,,= (^4^). It is p = q. Normally, these will all be positive. The absence described by the Lagrangian from (57) of any terms involving t?, but not d*, is indicative of the fact that the fields t? contain the L=^F"'(dyA,dl,A,)+\F^Fll, F"(a^aA«A(lXA,)+iF'"Fi„ massless excitations required by the Goldstone theorem. Now let us consider the coupling of the currents (44) + $»*• (drf+ek^Xt+ieArf) or (56) to a set of g gauge fields AAf. Thus, we now take +4*(d^+eAnXfieArf*) L=\Fir{dJi\dlA\tABCABtiA(!,} +lFA»>FAllv+P«dflpipi'p,
while the effect on 6" may most easily be expressed in the form d &V= #<,* l(A>)\(e*')cAy\A. (59)
which is positive definite because of the antisymmetry of the matrices Ta. We may summarize the situation as follows. Before introducing the gaugevector fields, we have r massless scalar fields which may be placed in onetoone correspondence with the broken components of the symmetry group, and n— r=v massive scalar fields. When the vector fields are coupled in, the v massive scalar fields remain, but the massless scalar fields combine with r of the vector fields to yield r massive vector fields. We are left finally with g—r massless vector fields corresponding to the unbroken components of the symmetry group. Thus, in order to avoid the appearance of any massless particles, it is necessary to choose a representation for which r=g; or, in other words, for which the subgroup G, of elements of G leaving y invariant is trivial or at most a discrete group. V. A SIMPLE MODEL
y*hV&*t;\f\).
V(p)
(64)
+*V<vr iCR^A'np":^} (X1) \j+<V(Ain(<r''0VlV (60)
This is invariant under infinitesimal transformations of the form S(4 = ieSA{14e5uX?i, 5 ^ = ie5X^fe5a>X(JP, lAy.— — d,,5X, 5F„„=0, SA^eScoXA,,— dpdio, 5F^=e5o)XF(,,,
We shall work implicitly in the Coulomb gauge, but will not explicitly indicate the Lagrange multiplier term
R. Hagen for commenting on dimensional (fundamental) representation must be an earlier version of the manuscript. secondly. merely by writing down the Lagrangian (66). and a single invariant ( r = l ) . where tj=(t<ik) = (eujk) (65) and P=2»»(p 1 +H») I and we may choose. and to reach identical conclusions. genergroup G. It is perfectly possible to describe it without ever introducing the notion of symmetry breaking. For example. the only two ACKNOWLEDGMENTS irreducible representations for which Gv is not trivial are the onedimensional identity representation and the I am indebted to Dr.£ M dpBJ+i&'G.. for SU(2). determined by the second derivatives of V. VI. and (r/1±r)2)2.. CONCLUSIONS In this paper we have tried to establish two main points: Firstly.study. We would also discover a pair of scalar particles forming an apparently incomplete multiplet under the group describing this symmetry. Guralnik for numerous threedimensional adjoint representation. except for a few representations cor. to set Pi U Pi= 0 . In particular it is not so easy to give the vector responding to small values of n. C.type. and to Dr. Provided The condition that there be no massless particles is also that we choose the form of V so that none of these the condition that no components of the symmetry quantities vanish. . There is just one unbroken component. (66) +h{pi+Pi)K&B#)(B. and. It therefore has four real components. Suppose we construct such a model for any SU(3).bosons a reasonable mass as to give them some nonzero variants constructible from <j> will be exactly v = n—g:mass. we would discover the existence of a set of four vector bosons with different masses but whose interactions exhibited a remarkable degree of symmetry. ] We note certain characteristic features of our model. Indeed if the physical world were really described by this model. it is (66) rather than (64) to which we should be led by experiment. [The two. Experimentally.22. For each unbroken component there remains a massless vector field. the photon. and one known massless the adjoint representation. that it is possible to handle the problem of symmetrybreaking consistently in the Lorentz gauge If as well as in the Coulomb gauge. S. and 02 = P2 oJ L= §G"'(d. Then. that in the case of nonthen we have in lowestorder four vector particles of Abelian gauge groups (as well as in the Abelian. it does at least seem worthy of further In that case the model will be completely free of mass. This is of course pretheory. However. for example. rents associated with a broken symmetry can serve to in addition to the two scalar particles whose masses are eliminate massless particles completely from the theory.j. Considerable difficulties still face any theory of this forming according to some other specified representation. G.vector boson. lo Then the transformation (61) leads to the equivalent Lagrangian + Pl"^pi+P2"^p2— hpi"Pl„— p2"P2^ FQ(P12+P22)ipi2P22) complex. less particles. no zeromass particles remain in the remain unbroken.Btr) +i(pip^(B"+Bf)(B„+B.discussions. (w = 4). It should be remarked that this model is in no sense cisely the physical situation in regard to groups like unusual. We may now write l = e"V'V. In such circumstances it would surely be regarded as a considerable advance if we could recast the theory into a form described by the symmetric Lagrangian (64). the number of in.case) masses the introduction of gaugevector fields coupled to cur7ji2. The only advantage of (64) is that it is easier to understand the appearance of an exact symmetry than of an approximate one. and n scalar fields <f> trans. with g vector fields transforming according to ated by the electric charge.172 155 SYMMETRY BREAKING IN NONABELIAN GAUGE THEORIES 1561 which belong to the group t/(l)X0(3)~Z7(l)X5[/(2) = 1/(2).ll)].
practical reason for making these calculations. FEYNMAN (Received July 3. but because I wish to limit my subject. And so I made * Based on a taperecording of Professor Feynman's lecture at the Conference on Relativistic of Gravitation. The effect of gravitation on the hydrogen atom is to shift the phase by 43 seconds of phase in every hundred times the lifetime of the universe! An atom made purely by gravitation. I am investigating this subject despite the real difficulty that there are no experiments. (657) Theories . as far as quantum effects are concerned let us consider the effect of the gravitational attraction between an electron and a proton in a hydrogen atom. so that we can have as many terms as we want. and therefore it is legitimate to analyze everything in perturbation approximation. Weber. has a Bohr orbit of 10 8 light years. which means that we can go to ten to the minus twohundred and something rydbergs. Therefore there is so real challenge to compute true. Jablonna. the correct problem is what determines the size of gravitation? But since I am among equally irrational men I won't be criticized I hope for the fact that there is no possible. an energy. Wheeler and other such things. P . — Ed. physical situations. so it's hard to explain why you do any of it.Vol. because the dimensions are so peculiar. 6 (12) QUANTUM THEORY OF GRAVITATION* BY R. XXIV (1963) ACTA PHYS1CA POLONICA Fasc. and I will carry out the perturbation approximation as far as I can in every direction. for example. 1963) My subject is the quantum theory of gravitation. in the absurd creations of Prof. July. let us say two neutrons held together by gravitation. It is therefore clear that the pi oblem we are working on is not the correct problem. I am limiting myself to not discussing the questions of quantum geometry nor what happens when the fields are of very shortwave length. This irrationality is shown also in the strange gadgets of Prof. My interest in it is primarily in the relation of one part of nature to another.7 0 rydbergs. I suppose that no wave lengths are shorter than onemillionth of the Compton wave length of a proton. it changes the energy a little bit. The energy of this system is 10 . not that I believe that gravitation is incapable of solving the problems that we have in the present theory. of the order 10~ 120 . Changing the energy of a quantum system means that the phase of the wave function is slowly shifted relative to what it would have been were no perturbation present. 1962. I am not trying to discuss any problems which we don't already have in present q u a n t u m field theory of other fields. There's a certain irrationality to any work in gravitation. I wish to discuss here the possibility of calculating the Lamb correction to this thing.
. g is written as flat space as if there were no gravity plus x times k „ where x is the square root of the gravitational constant.. I imagined that there were a lot of experiments and that the gravitational constant was more like the electrical constant and that they were comin°u p with data on the various gravitating atoms. and that it was a challenge to calculate whether the theory agreed with the data. Then.m 2 Vg <p2) ch (1) Substituting and expanding. it's immediately remarked that the theory is nonlinear. y . there are terms which are quadratic in h. I was looking entirely for unfamiliar (that is.v'P„u(p.a—2hlia3ahfiaia)+— + x f yif. is symmetric. where it is allowed (it is especially more allowed in gravity where the coupling constant is smaller).698 believe that there were experiments. I can only do one example at a time. I started with the Lagrangian of Einstein for the interacting field of gravity and I had to make some definition for the matter since I'm dealing with real bodies and make u p my mind what the matter was made of. then.v—rn2—hoa(p2} + a I "hhh"+>c° I "hhycp" + etc. not a question. I took spin zero matter. and then later I would check whether the results that I have depend on the specific choice or they are more powerfid. For example. I thought that this would be very easy and I'd just go ahead and do it. Also I decided not to investigate what I would call familiar difficulties. J2 = ^ j R Yg dx + \ J (fa £ " Y . This is not at all an unfamiliar difficulty. it's not something to be investigated unless it produces any specific difficulties associated with gravitation. one gets a big mess. the theory. if this theory diverges. So that in each case I gave myself a specific physical problem. In short. and so forth. . The quantum electrodynamics diverges. just as we do in quantum electrodynamics. unless that question was necessary to the solution of the physical problem. unfamiliar to meson physics) difficulties. of the spin 1/2 particles interacting with the electromagnetic field has a coupling term yAip which involves three fields and is therefore nonlinear. and I went to a harder one imagining the experimenters were getting into more and more complicated situations. what happens in a quantized geometry. which is outlined here. that's not a new thing at all. then I was finished. if I could do one. for example.. then there are terms which are quadratic . if this is substituted in the Lagrangian. Now. how do you define an energy tensor etc. since I'm going to make a perturbation theory. # J2 = = xf[V) we get I {<p%— m2q>2)dT + I (hfir. (2) First. so please appreciate that the plan of the attack is a succession of increasingly complex physical problems. and here's what I planned. and simplifying the results by a notation (a bar over a tensor means notice that if x/n.ahli.
analogous to the Lorentz condition A —0 in quantum electrodynamics. we take the variation of this with respect to /?. then there are terms which involve two h's and two <p's. and so on and so on with more and moie complicated terms. Making the simplest choice which I know. because of the invariance of the equations. of course. times 1/k2 in Fourier series. the entire system is invariant: h]„ = hliv + 2„ f . But as you all know it is singular. and to use the usual methods of calculation of the quantum theory. in particular). So I have to solve this equation (3). The first two terms are considered as the free Lagrangian of the gravitational field and of the matter. because two terms in (3) fall away and all we have is that the d'Alemberian of h is equal to 5. if you add to A a gradient plus more. and p.oom2<P = %{<P'h)(3) (4) We will speak in the following way: (3) is a wave equation. Now. (5) where £ is arbitrary. from the first term we produce a certain combination of second derivatives. and v should be made symmetric in all these equations. of which 5 „ is the source. As a consequence of this same invariance in the complet Lagrangian one can show that the source S „ must have zero divergence Sfiv> „ = 0. In fact equations (3) would not be consistent without this condition as can be seen by barring both sides and taking the divergence — the left side vanishes identically. which I will write lupq> (I have written that one out. This is the analogue of the equation in electricity that says that the field is 1/k2 times the current. and the plan in electricity leads to the following suggestion here: that if you have a source. I just took a guess that I use the same plan as I do in electricity. in the same way that the Maxwell equations cannot be solved to get a unique vector potential — so these can't be solved and we can't get a unique propagator. And the same with the cp. wave vector.. Then this equation (3) is very simple. This is four conditions and I have free the four variables  / ( that I can adjust to make the condition satisfied by h' „. The problem is to solve those equations in succession. the spin zero meson field variable. just like (4) is the wave equation of which % is the source. is the following. here is a term with two <p's and one h. — Kp. the entire Lagrangian in the beginning was invariant under a complicated transformation of g. which in the form of h. In the method of quantum field . and on the other side a mess involving higher orders than first. K'.OOKv. there are terms with three h's. where Jc2 is the square of the frequency. for example. Therefore the generating field from a source SMV will equal the 5 . Inasmuch as I wanted to get into the minimum of difficulties. then there are terms which are more complicated than quadratic.699 in q>. (f) <P. + 2 / v fB> „ + f„ / v > „. the time part might be called the frequency co.OV = "V (h. I make choice h^ = 0. Now we look first at what we would want to solve problem classically. the space part k. But because of the invariance under the transformation some arbitrary choice of a condition on h v can be made. you divide by the operator on the left side of (3) in momentum space to get the propagator field.Of.
There is a pole in the last term when co = k. is S/lV h/iv. we would take the first term. you have a source which generates something. so there's an instantaneous ljr interaction between masses. After some rearranging there results P" = U2 L"^44 ^ii\ + i"~2 l'. And that take care of gravitation to a very high accuracy. Let index 4 represent the time. so that. which means that there are two kinds of waves. and 1/k2 means 1/r in space.700 theory. However.44('>ll + $22) + 5 4 4 ( 5 U + S 22 ) + — '"• ~h2 + S 4 3 543 —45 41 S41— 4S 42 542] + rg . Radiation of free gravitons corresponds to the situation that there is a pole in the propagator. many of the terms involving number 3 component of 5 can be replaced by terms in number 4 components. it can be put directly into a Schrodinger equation. Newton's law. I say. and thai may interact later with something else. so that 1 and 2 are transverse.. is the product of two terms. The residue of the pole. of two particles. = 0 becomes coSir = kS3>.. So if we really wanted to do problems with atoms that were held together gravitationally it would be very easy. ^ [(Sn—5 2 2 ) (Su—S22) + 45 12 Si2] • (7) There is a singular point in the last term when co = k. in the Fourier transform an interaction instantaneous in time. without a quantized field theory at all. where k is the magnitude of k. The condition \Sfl. the iteraction. But the k 2 — a>2+it in gravitation comes even later and is a much smallei term which involves velocities to the fourth). and 3 the direction of k. 5 4 4 is the energy density. of course. . which means that the wave number and the frequency are related as for a mass zero particle. so this first term represents the two energy densities interacting with no co dependence which means. and to be precise we put in the + ie as is wellknown from electrodynamics. analogous to the e2/r term for electrical interaction. to take the very simplest example of two interacting systems. of course. the result would be the following: h would be generated by S . so we would get for the interaction of two systems. To understand the theory better and to see how far we already arrived we expand it out in components. the momentum itself. Being instantaneous. let's say S and S'. one generated by Sn— S22 and the other generated by S12. we see. In the next term there's another instantaneous term which says that Newton's mass law should be corrected by some other components analogous to a kind of magnetic interaction (not quite analogous because the magnetic interaction in electricity already involves a k 2 —co 2 +ie propagator rather than just k 2 . one source may create a potential which acts on another source. the fundamental interaction that we investigate This represents the law of gravitational interaction expressed by means of an interchange of a virtual graviton.. You note that in the first two terms instead of one over a fourdimensional co 2 —k 2 we have here just 1/k2. and possibly even the second as the interaction. which come from the last term. and then it would interact with 5. and so we have two lands of trans . for still higher accuracy we have to do the radiative corrections. Using this. So..
that is A a(3 = eafj elqx. or the analogue rather. and I shall just illustrate how that's done in one example. which must exist. It is done by looking at the nonquadratic terms in the Lagrangian I've written one out completely. so we immediately see in that term a two particle interaction through a graviton (see Fig. it is very complicated. A term like hhh implies a definite formula for the interaction of three gravitons. that is there are two polarization states for the graviton. 1). of the Compton effect. of course. including its own. I've already indicated that the physics of direct interactions is sensible. I just wanted to illustrate that the propagator (6) of quantum mechanics and all that we know about the classical situation are in evident coincidence. The rules of the quantum mechanics for writting this thing are to look at the h and two <p's: one <p each refers to the in and out particle. but you can read it right off directly by substituting momenta for the gradients. we can write everything. the Fourier expansion of this term gives the amplitude for the coupling of two particles to a graviton pip* ef. The amplitude . Everything is clear directly from the expression (7). natural. (8) So this is a coupling of matter to gravity.701 verse polarized waves. it is first order. And we can immediately read off the answer for the interaction this way: if the p1 and p2 are the momenta of the particles and q the momentum of the graviton. we compute the Compton effect. we have the couplings. The linear combination Su— 5 2 2 ± 2sS 12 vary with angle 0 of rotation in the 1—2 plane as e±2'° so the gravitaton has spin 2. beside the propagator we need to know just what the junctions are. To take just one example. and the one h corresponds to the graviton. but the point I'm trying to make is that there is no mystery about what to write down —• everything is perfectly clear. That such a term exists is. and then there are higher terms. because gravity interacts with any kind of energy. in which a graviton comes in and out on a particle. from the Lagrangian. We have the propagator. so this is the scattering of a graviton in a gravitational field. and ea/3 is the polarization tensor of the plane wave representing the graviton.v— ^ m2eaa. This one has an h and two <p's in the Lagrangian (2). and I won't write it down. and I go ahead now to compute a number of other things. component ± 2 along direction of polarization. in other words just what the 5's are for a particular problem. So that everything is directly readable and all we have to do is proceed to find out if we get a sensible physics. In order to proceed to make specific calculations by means of diagrams. so if it interacts with an objectparticles it will interact with gravitons.
2. If I put the gradient qa£p in for e^ on this term D. which is the Fourier transform of the equation (4) which is the propagation of the spin zero particle. Then you must add similar contributions ~jF^ ?« + .702 for this is a sum of terms corresponding to the diagrams of Fig. because you would have expected that the invariance law of the whole thing is more complicated. it should really be invariant under coordinate transformations and so on. i i ° \Pa Px ^CTr "o. as far as the matter line is concerned is that it is acted on in first order by a resultant field e^0 £a qav +  qha e^ £a which is just the last two terms in (5). you see asymptotically all you have to do is worry about the second term. that it is simple is what is interesting. The amplidute for the first diagram of Fig. the last two in h's times f's are in fact generated by the last diagram. But those terms have been included. what this means is if I put for the incoming wave a pure gradient. I get a coupling between £ and the other field e"aB because of the three graviton coupling. and it is. In short. We multiply these together. including the last two terms in (5). there should be no change in the physics. One adds the four diagrams together and gets an answer for the Compton effect. It is rather simple. Fig. and quite interesting. in this problem we used a certain wave e^g for the incoming graviton number " a " say. The amplitude for this particular process is what I call gaugeinvariant. Then there is another coupling of the same form as (8). The result.. and qa is the momentum of the graviton.p where £ is arbitrary. 2D. but what it corresponds to here is the analogue of gauge invariance. 2 is the coupling (8) times the propagator for the intermediate meson which reads (p2—m2)^1. when I put a gradient in here for this one. At first sight this is somewhat puzzling. But the thing I would like to emphasize is this. 2 The third one comes in because there are terms with two h's and two q>'s in the Lagrangian. because the labour is fantastic in all these tilings. which I seem to have omitted. I should get zero. that you can add to the potential a gradient (see (5)). the question is could we use a different one? According to the theory. or coordinatetransforming invariant.^oa^ 2 where we should substitute p = p2+qb from the other diagrams. to get the amplitude for that diagram 2 6 PuPv e. Fig.l„ 1 1 _ 2\ ^ 2 etlfl m / j p"—m 2 0 = p1+qa.. The rule is that the field which acts on the . the amplitude should be unchanged. And therefore it should be that if I changed e^ of a particular graviton to eajS +qj.
The sum of all diagrams is zero. and therefore if I were to just calculate the diagrams of Fig. 3 c Now. If this " a " field is taken as a gradient field which o p e r a t e s e v e r y w h e r e o n e v e r y t h i n g i n t h e d i a g r a m it should give zero. this is not a disease — there is a physical reason.) Sp'S. now. The reason can be seen as follows: If this b had a source let me modify my diagrams to show the source of b. source of o Ji ^ \ ^ . in which not all momenta of the problem are defined. o n l y if b is a f r e e g r a v i t o n w i t h n o s o u r c e . That will turn out to be an important point to emphasize. {E. 3a. There is no way to avoid this by changing any propagators. one of which looks like Fig. there's nothing new. 2 I've already calculated all the corrections. I have done a lot of problems like this. among all the diagrams where there is a source. The result is zero if I use a gradient for a. Let's call a the one graviton coming in and b the other one in every diagram. there's another type of diagram. the generator and all the necessary nonlinear modifications if I take all the diagrams into account. That is unlike electrodynamics. where the field b could have been a n y potential at all and adding a gradient to a would have made no difference. but here in Fig. the invariance need be checked only for a pure gradient added to an incoming wave. It takes care of the nonlinearities by calculating them through the interaction. but the sum of those like Fig. 3b etc. But we forgot something. or rings. I said. like Fig. Fig. asymptotically far away if I include all kinds of diagrams such as D. In short. However. there's also these of type 3c. in the sense that no apparent difficulties arise. but must be getting ready to cancel the terms from the likes of 3c when I do it right. the next step is to take situations in which we have what we call closed loops. But in gravity. which is a solution of the free wave equation. if the " a " is supposed to act on everything. without closed loops but I won't bore you with all the problems and answers. If I add a gradient. if b isn't a free wave. 3c.g.703 matter itself must be invariant the way described by (5). I mean nothing interesting. but it had a source. I would like. ^^_. 2a becomes Fig. 2 without those of type 3c is not zero. 2 and forget about the source of b and then put a gradient in for " a " the result cannot be zero. so we add onto each b line a matter line at the end. the situation is this. it must be that b is a pure wave. or a pure potential. In other words. Let me just men . to emphasize one more point that is very important for our later discussion. suppose some other matter particle made the b. in which the " a " itself acts on the source of b and then b comes over to interact with the original matter. the result was zero. that is if it is either really an honest graviton with (qb)2 = 0.source of b a ^~^ a b Fig. the reason is very simple. or circuits.
In order to get it gaugeinvariant. because in spite of getting around all those things the result is nevertheless definitely incorrect. a graviton gets split into two gravitons and then come back — these are only the beginning of a whole series of frighteninglooking pictures. This is not meant as a grand discovery. Shortage of time does'nt permit me to explain the way I got around all those things. I had to do a lot of pushing and pulling. that's the difference. I can prove that no matter how complicated the problem is. I did it in a straightforward way. It's gaugeinvariant. and so on. so I tried to get more tests for it. it's perfectly O.K. and so I 'm really saying there are no difficulties in the classical gravitation theory. by the name of Yura. you've been worrying about all these difficulties that I say don't exist. 4 or. for instance. looking. but I don't believe in it. and everything is all right. let's say the interaction of two particles in a higher order. . and a student of mine. in which every momentum is determined. following all the rules. tested to see if it was unitary.704 tion something. but it is definitely incorrect. and by a mathematical highclass elegant technique — I can do high class mathematics too. The reason I knew it was incorrect is the following. if you take it in the order in which there are no rings. which correspond to the problem of calculating the Lamb shift. 4 in which time runs vertically. and I was rather suspicious. to produce another pair. the invariance is satisfied. I've analyzed this method both by doing a number of problems. Let's suppose for simplicity that in the final state the pair is in the same state as before. When I tried to do this. a problem of Fig. or the radiative corrections to the hydrogen atom. and we are calculating second order corrections to that problem. a problem which gives the same diagrams but in which time is running horizontally. there are no difficulties. because after all. I had some difficulties. One that looks like this in which two gravitons are exchanged. Fig. the thing didn't look gauge invariant but that had to do with the way I was making the cutoffs. and what that means is the following: Let me take instead of this scattering problem. which is the annihilation of a pair. I figured that maybe somebody else could do it another way or something. and I got the feeling that the thing might not be unique. I emphasize that this contains all the classical cases. but only for you to get an idea of the calibration — what I mean by difficulties! If we take the next case. I have to check it in a problem. then you get diagrams of which I'll only begin to write a few of them. putting in the propagator 1/k2. because the stuff is infinite. the system is independent of what choice I made of gauge and of the propagator I made in the beginning.
as I said.705 Then. in the past. so that you can express any diagram with circuits completely in terms of diagrams with trees and with all momenta for tree diagrams in physically attainable regions and on the mass shell. incoming waves and in the future. annihilate. In a ring like Fig. that's a certain propagation function which you might call K+(i. which relate closed loop diagrams and diagrams without closed loop diagrams (I shall call the latter diagrams "trees"). Somethin'gs the matter. 2). they annihilate and recreate themselves. It represents. 2. Therefore. what is the rate at which a particle and antiparticle annihilate into two gravitons ? And this is very easy to calculate (same set of diagrams as Fig. for instance there is a solution which is purely retarded. Why does the probability decrease in time? Because there's another possibility. This made me investigate the entire subject in great detail to find out what the trouble is. 2. in spacetime. 4). with the ie and so on. which as far as I know are new. I proved that if you have a diagram with rings in it there are enough theorems altogether. I discovered a number of theorems. The demonstration is remarkably easy. checked whether this rate agrees with the rate at which the probability of the two particles staying the same decreases (imaginary part of Fig. However. First. it is necessary that this decay rate of the closed loop diagrams in Fig. but I found a whole lot of other ones. these two objects could come together. It is rather interesting. and produce a real pair of gravitons. which I'll call iCret and which exists only inside the future lightcone. . it represents outgoing waves. For example. Namely. without looking at the closed loop diagrams. There are other solutions. and it does not check. 3)X+(3. if you have two Green's functions for the same equation they must differ by some solution of the homogeneous equation. which means that the probability of being in a state must decrease with time. There are several ways of demonstrating it. some of them belong to the gravitons and some are propagators for the particles and so on). adding all these diagrams gives the amplitude that if you have a pair. say Kx. 4)X+(4. I calculated this rate from Fig. 2)K+(2. 4 in a ring we have a product like this: K+(l. 4 that I obtain by directly finding the imaginary part of the sum agrees with another thing I can calculate independently. so you have waves come in and out. 1. is one connection between a closed loop diagram and a tree. So let me just tell you a little bit about this theorem. The amplitude to remain in the same state for a time T in general is of the form you see that the imaginary part of the phase goes as e . Things propagate from one place to another. 4a we have a whole product of these K+'s. particle and antiparticle. only turned on its side). Now. a function of two positions. That means Kx is a solution of the free wave equation and K+ = Kzet + Kx. as usually represented. As a matter of fact. the wave equation I mean. in other words it's the amplitude that the pair is still in the same state as a function of time. and that's the conventional propagator. for four points 1. with amplitude 1/k2. 2. 2. it is a special solution. and this gives me more tests on my machinery. When translated into space. this is only a solution of the propagators's equation. The unitarity relation which I have just been describing. 1) (all iCs are not the same. I discovered in the process two things. namely. 3. as you all know. which gives other rules. I'll only chose one.
) That is we want only —S^ —S2S2 instead of the sum over four. closed loops can be represented as integrals over tree diagrams. It ins't. it is not transverse. First I note that if I put KKt for every K in a ring. Since the sum is zero. say. Now make the substitution Knt = K+—Kx in (9). 2)? Then between 1. then I checked whether these diagrams of Fig. Now when I make q2 = 0 I have a free wave instead of arbitiary momentum. but it is the same idea. a typical propagator is a factor l/(k2—m2 +is). But I said that I could open every diagram. not the denominator. I was surprised I had never noticed this thing before. So Si—S3 is the response of the system to a gradient . What I want these free waves to represent in the gravity case are physical gravitons and not something wrong. I get zero iT r e t (l. how could it fail. But St—S3 is proprtional to q S^ (suppressing one index) because qi in this notation is the frequency and equals q3. You see the propagator that I wrote before. I'd like to show it. t2 > t3.706 But now let us see what happens if we were to replace one (or more) of these K+ by Kxi say X+(l. This s h o u l d be a real graviton or else there's going to be physical trouble. Well. which is the free wave coming in and out. you've broken the ring. and all I'm doing here is changing the rule about the poles and picking up an extra delta function d(k2— m2) as a consequence. Then 1 and 2 are transverse. the reason is this. 4)X r e t (4. (Please appreciate I removed one index — I can make it more elaborate. It has to do with the numerator. if we assume the 3direction is the direction of the momentum. you've got an open diagram. You get either all K+ in each factor. and the process is changed to one in which there is a forward scattering of an extra particle. it is Si—S3 times 5^ + S3 plus 5 4 + 5 3 times S^—S^. you say. where 4 is the time and 3 is the direction. on the mass shell and perfectly honest. And it does not work in the gravity case. as it turns out. Now what about this extra term S^— S3S'a? Well. although it is of zero momentum. because Kx is free wave solution. 2 we have just free particles. is being replaced by S/Jvd(q2)S'/iV. Therefore if we replace one of K+ by Kx then that particular line is opened. not with the correct polarization to be free gravitons. 3)£ r e t (3. 2)/<Tret(2. 1) = 0 (9) for to be non zero tx must be greater than t2. such that on opening a virtual graviton line the tree would correspond to forward scattering of free graviton waves. t3 > £4 and i 4 > tx which is impossible. which is the closed loop we want. This combination S^S'^. and those are the only two we want. 2) is Kx(\. which was SMV times l/(k2 + ie) times 5^„. there's a fake particle that belongs to this propagator that has to be integrated over. and therefore perfectly definite and unique to calculate. 4 when opened into trees agreed with the theorem. It does not make any difference in understanding the point so forget one index in S^ — it's a lot of extra work to carry the other index so just imagine there's one index: SflSftd(q2). but it's a free diagram — it is now a tree. and this means it's an integral over all real momenta of free particles. the +is due to the poles. but. is 5 4 5 4 —5 3 53—5^j—5 2 52. But. I mean I hoped that the theorem proved for other meson theories would agree in principle for the gravity case. of momentum of the fourvector q. They do represent waves of q2 = 0 of course. which are represented by tree diagrams. after you just demonstrated that it ought to work? The reason it fails is the following: This argument has to do with the position of the poles in the propagators. or at least one Kx.
more wellknown to meson physicists. then there are many things you don't worry about. So at least there is one good thing: gravity isn't alone in this difficulty. Such a wave could only be generated by an artificial source here of some silly kind. if I ever cure that one. At the suggestion of GellMann I looked at the theory of YangMills with zero mass. If the beam is not exactly that of a free wave. and which is not a free wave — and therefore you do not get zero and should not get zero. with the idea. for gravity. that's not the cause of the trouble. You see. that's like the gravitation theory. It is a nonlinear theory. which has a kind of gauge group and everything the same. And therefore in meson theory it was not strictly unknown difficulty. perfectly transverse and everything. for the reason outlined previously. Because I can demonstrate one thing. And the second great advantage was that the YangMills theory is enormously easier to compute with than the gravity theory. But it ins't. and the YangMills theory with zero mass obviously does not exist. and therefore I continued most of my investigations on the YangMills theory. line for line it's a translation like music transcribed to a different . it is not a correct kind of graviton. we have shown (Si—S3)/(S'i+S'3) = 0 and this should be accounted for by purely transverse wave contributions.707 potential. The main reason is if you have two examples of the same disease. which we proved was zero in our invariance discussion. You see. There is another theory. it's not a transverse wave. it's the isotopic spin rotation group that's the source of everything. Incidentally I investigated further and discovered another very interesting point. it is not a free wave. it is a special theory that has never been investigated in great detail. So they didn't take the case of zero mass and investigate it carefully. That's one advantage — it limits the number of possibilities. This observation that YangMills was also in trouble was of very great advantage to me. in front of the second derivatives of g v in the Lagrangian there are other g's. and so forth. and found exactly the same difficulty. and it isn't because the proof that the response to a gradient potential is zero required that the other particle that was interacting was an honest free graviton. the only way you can get a polarization 4 + 5 going in the 4—3 direction is to have what I call longitudinal response. for several reasons. even the another is a pure gradient. it would come out of nuclei right away. and something is fundamentally wrong. because a zero mass field would be obvious. Although this gradient for Si—Sz is what I want and I hoped it was going to be zero I forgot that the other end of it —• S4+S3 is a funny wave which is not a gradient. But this disease which I discovered here is a disease which exist in other theories. the sum of all the diagrams does not give zero. if there is something different in the two theories it is not caused by that. instead of the coordinate transformation group being the source of everything. the argument that the gradient has to be zero must fail. the field itself. And four plus three in 5^ + 5 3 is not honest — it's not transverse. it made everything much easier in trying to straighten out the troubles of the preceding paragraph. They had not noticed it because they're practical. and I take the one with zero mass. I'll turn around and cure the other. called the YangMills theory. I kept worrying something was going to happen from that. It is very analogous to gravitation. For example. When there's an artificial source for one graviton. because it should have been noticed by meson physicists who had been fooling around the YangMills theory. and so on. Therefore. In the YangMills theory this is not so.
For instance the tree diagram of Fig. and is one of the class of relations I was talking about. for example. nor is any combination of them except the sum of all four. all of them that belong together in a given order. I mean the algebra of all the terms coming in. second. everything has its analogue precisely. lots of trees. So that's the plan. to calculate the diagram in the case of gravitation I tried again and again and was never able to do it. then all you have to do is to say that the closed loop diagram is the sum of the corresponding tree diagrams. being superseded. The business of gathering the tree diagrams together in bunches representing all diagrams for complet processes is important. 2A is not. If every closed ring diagram can be expressed as trees. 2 is not gaugeinvariant. it's completely unique as to what the answer is. to give you some idea of the difference in order to calculate this diagram Fig. for only such a complet set is gauge invariant. that it will be gauge invariant. that it should be. Each one of the four diagrams of Fig. and if trees produce no trouble and can be computed. The advantage of this is. a given t r e e is not gauge i n v a r i a n t . so there's no difficulty. So the thing is the following. one more point. This then serves as the definition of how to calculate closedloop diagrams. In fact. the old rules. but the algebra of the thing was done on a machine by John Matthews. but. not just some particular tree diagram but the complet set for some process. and used it in reverse. all the diagrams of fourth order. the plan requires. take only real transverse graviton to represent that term. Well. of which Fig. 4 illustrates three. you see. and it was finally put on a computing machine—I don't mean the arithmetic. 5 the trees into baskets again. just the algebra. Then I must gather Fig. now you have the difficulty. And third. so that each basket contains the total of all of the diagrams of some specific process (for example the four diagrams of Fig. how do you cure it? Well I tried the following idea: I assumed the tree theorem to be true. involving a propagator ljk2+ie etc. The question is: Will any odd tree dia . it will be unitary. so it is a very good example to work with. I did the integrals myself later. Now. Suppose I take all the processes. Incidentally.708 score. Finally in each tree diagram for which a graviton line has been opened. It's true that we proved here that every ring diagram can be broken up into a whole lot of trees. 2). I think it's historically interesting that it's the first problem in algebra that I know of that was done on a machine that has not been done by hand. 4b the YangMills case took me about a day. so I couldn't have done it by hand. I break the whole mess into trees. first. what then. there's no arbitrary fiddling around with different gauges and so forth. however. in the inside ring as there was before. because unitarity is a relation between a closed diagram and an open one.
There is no external graviton but there are two internal ones. it will Fig. 6 give the same answer as this unique. You might be interested in what the rule is for one ring. So I investigated that in detail. change the rule for integrating the closed rings. in Fig. In addition. in the case with one ring only. so another graviton comes in and is scattered forward.. Is it possible to go back and to find the rule by which you could have integrated the closed rings directly? In other words. from breaking matter lines we have terms for the forward scattering of an extra positron. 4 we have the rings for two particles scattering. one adds the forward scattering of every possible extra particle together. it took me quite a while to find.709 grams be left out or can they all be gathered into processes? The question is: Can we express the closed ring diagrams for some process into a sum over various other processes of tree diagrams for these processes? Well. of course. For instance. State / the same state as g. but while that's happening another particle. I have other proofs for orther cases that are easy to understand. It's not necessary to do this. and so on. you find the processes that they correspond to. I am sure it can be done. the next thing that anybody would ask which is a natural. you also sometimes break two lines. for example a graviton scatters forward. with the new method. 5. its's quite beautiful. plus the forward scattering of an extra electron. In other words we do the forward scattering of an extra graviton. now we compute in the same order a new problem in which there are two particles scattering. it's the sum of several pieces: first it is the sum of all the processes which you get in the lower order. 6 with two extra particles scattering (here a graviton and electron) so it turns out you must now subtract all the diagrams with two extra particles of all kinds scattering. interesting thing to ask. Some of the diagrams for this are illustrated in Fig. compute each process and add them together. And therefore the problem with one ring is fundamentally solved. absolute. but it's of great interest to do this. in which you scatter one extra particle from the system. It turns out there are two changes that have to be made — it's a little hard to explain in . That is the first contribution. It's a nice rule. you express it in terms of open parts. because maybe I'll understand what I did wrong before. definite thing of the trees. I proved it can be done and I have done it and it's all fine. Now. and then you get diagrams like Fig. because. is this. I've defined everything. But when you break up the trees. Then add all diagrams with 3 extra particles scattering and so on. so that when you integrate them in a more natural fashion. because we say.
the result that you get by imagining that in the ring which involves only a graviton going around. unfortunately. In the meantime. You say to me what is this. But then secondly. we are still not sure. therefore. if it was not irrational before. there's a field variation over which you're integrating. when I made it gauge invariant. and I therefore do not claim that this method of quantization can be obviously and evidently carried on to the next order. Let me write what it really is so that there's no trouble. It's a vector particle. what happens when there are two or more loops? Since I only got this completely straightened out a week before I came here. In short. artificially coupled to the external field. these are like tensors in the g world. of the radiative corrections to the radiative corrections to the Lamb shift. no matter how absurd and irrational and academic it looks. and therefore it's clear that my answers are. and there's a g — which is the representative of all the outside disturbances which can be summarized as being an effective external field g. I'll try to explain the other. if it has physical interest and is thought about carefully enough.710 terms of the gravitation of which I'll only tell about one. I don't understand them at the moment. there seems to be some loose trees. Because I have to explain in general what I'm doing when I do a ring. I can't retire on the claim that the number is so small and that the thing is now really irrational. I can't understand it. and secondly I have incidentally discovered a treering connection which is of very great interest and importance in the meson theories and so on. an artificial. instead you calculate with a different particle going around. we all realize that no matter how small a thing is. And so you add to the complicated Lagrangian that you get in the ordinary way an extra term. And so I'm stuck to have to continue this investigation. That's the first thing. the uncertainty lies in energies of the order of magnitude of 10_25° rydbergs. you're bound to think of something that's good for something else. as far as your g'space is concerned. extra trees. Now. and all that's necessary. there's a g in it and an H in it? Yes. I also discovered in the process that the trouble is present in the YangMills theory.gauge invariant or coordinate transformable. dopey particle is coupled to it. and say: for all practical purposes everything is all right. I haven't had time to inwestigate the case of 2 or more loops to my own satisfaction. which I call H. I can therefore relax from the problem. unfortunately. Because. but it might cause some confusion. you must subtract from the answer. The preliminary investigations that I have made do not indicate that it's going to be possible so easily gather the things into the right barrels. But are also quantummechanically satisfactory in the sense that they are unitary. In doing a ring. the next question is. so designed as to correct the error in this one. . The forms are evidently invariant. which makes it no longer singular. I found it out by trial and error before. when you gather the trees into processes. although I could retire from the field and leave you experts who are used to working in gravitation to worry about this matter. It's surprising. Most what it corresponds to is this: first you subtract from the Lagrangian this In that way the equation of motion that results is nonsingular any more. and of course you all appreciate that this is the secret reason for doing any work. Well.
things like this that you see labelled as loops are very typical quantummechanical things. It could be done I'm sure. then they have to be particles of specified properties. I checked it by the mathmatics. and it gives the same answer in the limit as the corresponding classical problem. but even here you see a tendency to write things with the right derivatives. in fact when I work out the fields and I don't say in what order I ' m working. And I don't notice in the morass of things that something. the other is to check it by calculations. as a nonexpert. Or the gravitational radiation when two stars — excuse me. n o . in several orders and so on. M a i l e r : But you say you are not sure it is renormalizable. No. such as the rate of radiation from a double star held together by quantummechanical force. I don't notice when I ' m doing a path integral over an infinite number of variables that the Lagrangian does not depend upon one of them. I'm lousy at proving things — I always make a mistake. R o s e n : I'm not sure of this. or that the space is curved at infinity. But I've done two things. in which you have matter out to infinity. The classical limit of this theory that I'm working on now is a nonlinear theory exactly the same as the Einsteinian equations. I have to do it i n an abstract manner which includes any number of gravitons. there is still a classical electrodynamics.711 DISCUSSION M 0 l l e r : May I. a little limit or sign. there's no question that the thing is the Einsteinian theory. b u t I haven't investigated it. There's no question about that. Is this theory really Einstein's theory of gravitation in the sense that if you would have here many gravitons the equations would go over into the usual field equations of Einstein? F e y n m a n : Absolutely. So it's a lot of work in these things. The infinities come in different places. to any order you want (not for stars. If you do a real problem with real physical things in in then I'm sure we have the right method that belongs to the gravity theory. It can't take care of the cosmological problem. the integral is infinite and I've got a ratio of two infinities and I could get a different answer. you see. M a i l e r : In the limit of large number of gravitons this would not matter? F e y n m a n : Well. ask. It's not a related problem. F e y n m a n : I'm not sure. that the forms of the mathematical equations are the same. but I have the impression that because of the nonlinearity of the Einstein equations there exists a difficulty of the . a n d then the formulas are definitely related to the general theory's formulas. and then I checked it by doing a considerable number of problems in quantum mechanics. goes wrong. I used as a background a flat one way out at infinity. I have mathematically proven to myself so many things that aren't true. and it's not got to do with the renormalizability of quantum electrodynamics. One thing is to prove it by equations. and I'm very poor at calculations — I always get the wrong answer. no. gauge invariant and everything. M o H e r : You are quite sure about it? F e y n m a n : Yes. you a very simple and perhaps foolish question. because obviously the rate of radiation of the gravity depends on the give of the starstides are produced). So I always have to check with calculations. not being one of the experts. and the invariance is the same. two particles — go by each other.
Yes? F e y n m a n : It was one of the examples. that the thing was emitted some time far in the past. There's only one ha>. and has not absolutely been going on forever. 2 (there is no 2D) is satisfied whether b is a free wave or not. that the corrections to the propagation of a graviton. then t h e r e are corrections for such a ring as this and so on. If the linear equations have a solution in the form of an infinite plane monochromatic wave. you do not. there's no meaning to the correction. this is the shadow of the phenomenon you're talking about. F e y n m a n : I tried that and it didn't go successfully. a term in the logarithm of q2. dependent on the logarithm of the momentum squared carried by the graviton and which would be infinite if it were really a zero m o m e n t u m graviton exactly. A single graviton is not the same thing as an infinite gravitational wave. because you get piling up of energies in space and the solution then diverges at infinity. I can still get a sensible answer. you might be able to get easily by always only starting out with vacuum diagrams and opening those successively. . and is going to be absorbed some time in the future. I take that into account by a series of corrections. because there's a limited energy in it. R o s e n : But you're using a momentum expansion which involves infinite waves. but second. a divergence as usual. have any bearing on the accuracy of this kind of calculation? F e y n m a n : No. like the diameter of the universe or something. Gauge invariance of diagrams such as Fig. And this is the correction for that. That is because photons are not the source of photons. F e y n m a n : Yes. I v a n e n k o : If I understood you correctly. there are corrections. they are uncharged. And so a free graviton just like that does not quite exist. Strictly we would have to work with wave packets. And these produce first. So it must be understood in this way. sir. If you have here a graviton coming in this way. but they can be of very large extent compared to the wave length of the gravitons. you had used in the initial presentation the transmutation of two particles into gravitons. A n d e r s o n : The other thing I would like to suggest is that in putting of things into baskets. which means that if this thing is absolutely a free plane wave. Then there's a very small coefficient in front of the logarithm and then for any reasonable q2. You see what happens if one calculates the corrections. there does not seem to correspond to that a more exact solution. A n d e r s o n : I'd like to ask if you get the same difficulty in the electromagnetic case that you did in the YangMills and gravitational cases? F e y n m a n : No. Could that.712 following kind.
if one puts numbers in this. but at least it can't be any bigger than this. This is of course too small to measure. I v a n e n k o : They increase very. It may be less than that. And then the crosssections of any of these processes are infinitesimal. it depends on the problem. which is not as fancy as the one you have just heard about. and it turns out that even in the linear approximation that one has the possibility of the graviton production by scattering of photons in a Coulomb field. one finds that the scattering crosssection of a galaxy due to a uniform magnetic field through it is 1028 cm2. you've got to go to ridiculous energies. and as has been pointed out many times. You can figure out the answer right away. W e b e r : I have a crosssection which may be a partial answer to Ivanenko's question. In fact. You still might not get somewhere. On the other hand. but at least it can't get up any worse than that. just take the energy that you are interested in. So you either have a ridiculously small effect or a ridiculous energy. which is 1018 BeV. So if E is the energy of some process. we considered the possibility of using this crosssection for a laboratory experiment in which one had a scatterer consisting. in which case the rate is similar to the rate of photon annihilation. perhaps in a preliminary manner. Also. but considering the interaction of photons and gravitons. square. This process was considered. because the crosssection might not go up that fast. that's an energy of the order 10~5 grams. so we turned around and calculated it classically. So I think that in order to get an appreciable effect. multiply by G and divide by he. Could I write it on the board? We have carried out a canonical quantization. So in order to make this thing to be of the order of 1 %. Now. This represents a conversion of photons into gravitons of about 1 part in 1016. it starts out so low that one has to go pretty far to get anywhere. We obtained this result by quantization. this is so because of the dimensions of G. very sharply with energy. then if you take GE2jhc you get an equivalent to this e2jhc. so it increases sharply in contrast to the electromagnetic transmutation of an electronpositron pair. a much larger number than the object that you talked about. say of a million gauss magnetic field over something like a cubic meter. by ourselves and by Prof. a result in total contradiction to what has appeared in the Russian literature. Yes.713 I v a n e n k o : Yes. And the scattering crosssection for this case turns out to be 8n2 times the constant of gravitation times the energy of the scatterer times the thickness of the scatterer in the direction of propagation of the photon through it divided by c4. which is 1/137 is nonexistent in gravitation. This turned out to be entirely impossible. then you're getting somewhere. Do you possess the effective crosssection? Can you indicate the effects for which highenergy processes play an important role? F e y n m a n : I never went to energies more than one billion billion BeV. at ordinary energies. And the distance that you have to go is involved in this thing—the thing that's the analogue of e2/hc in electricity. and noticed that it didn't have Planck's constant in it. if that becomes something. we need the GE2 to be of the order of he. Weber and Brill. because the radiation is quadrupole. F e y n m a n : It increases very sharply indeed. This assumes that all of the dimensions of the scatterer are large in comparison with the wave length of the photon. I ask you two questions. the theory of fluctuations shows that for a laboratory experiment involving the production of gravitons .
714 by scattering of photons in a Coulomb field, the scattered power has to be greater than twice the square root of kT times the photon power divided by the averaging time of the experiment. I believe that the incorrect results that have appeared in the literature have been due to the statement that AP has to be greater than kT over T; dimensionally these things are the same, but order of magnitudewise this kind of experiment for the scatterer of which I spoke requires something like 1060 watts. Maybe I can say something about this afternoon; I don't want to take any more time. De W i t t : I should like to ask Prof. Feynman the following questions. First, to give us a careful statement of the tree theorem; and then outline, if he can to a brief extent, the nature of the proof of the theorem for the oneloop case, whish I understand does work. And then, to also show in a little bit more detail the structure and nature of the fictitious particle needed if you want to renormalize everything directly with the loops. And if you like, do it for the YangMills, if things are prettier that way. F e y n m a n : I usually don't find that to go into the mathematical details of proofs in a large company is a very effective way to do anything; so, although that's the question that you asked me — I'd be glad to do it — I could instead of that give a more physical explanation of why there is such a theorem; how I thought of the theorem in the first place, and things of this nature; although I do have a proof — I'm not trying to cover up. D e W i t t r May we have a statement of the theorem first? F e y n m a n : That I do not have. I only have it for one loop, and for one loop the careful statement of the theorem is... — look, let me do it my way. First — let me tell you how I thought of this crazy thing. I was invited to Brussels to give a talk on electrodynamics —the 50th anniversary of the 1911 Solvay Conference on radiation. And I said I'd make believe I'm coming back, and I'm telling an imaginary audience of Einstein, Lorentz and so on what the answer was. In other words, there are going to be intelligent guys, and I'll tell them the answer. So I tried to explain quantum electrodynamics in a very elementary way, and started out to explain the selfenergy, like the hydrogen Lamb shift. How can you explain the hydrogen Lamb shift easily? It turns out you can't at all — they didn't even know there was an atomic nucleus. But, never mind. I thought of the following. I would explain to Lorentz that his idea that he mentioned in the conference, that classically the electromagnetic field could be represented by a lot of oscillators was correct. And that Planck's idea that the oscillators are quantized was correct, and that Lorentz's suggestion, which is also in that thing, that Planck should quantize the oscillators that the field is equivalent to, was right. And it was really amusing to discover that all that was in 1911. And that the paper in which Planck concludes that the energy of each oscillator was not nhco but (n\lj2)ha> which was also in that, was also right; and that this produced a difficulty, because each of the harmonic oscillators of Lorentz in each of the modes had a frequency of hco/2 which is an infinite amount of energy, because there are an infinite number of modes. And that that's a serious problem in quantum electrodynamics and the first one we have to remove. And the method we use to remove it is to simply redefine the energy so that we start from a different zero, because, of course, absolute energy doesn't mean anything. (In this gravitational context, absolute energy does mean something, but it's one of the technical points I can't discuss, which did" require a certain skill to get rid of, in making
191 715 a gravity theory; but never mind.) Now look — I make a little hole in the box and I let in a little bit of hydrogen gas from a reservoir; such a small amount of hydrogen gas, that the density is low enough that the index of refraction in space differs from one by an amount proportional to A, the number of atoms. With the index being somewhat changed, the frequency of all the normal modes is altered. Each normal mode has the same wavelenght as before, because it must fit into the box; but the frequencies are all altered. And therefore the hco's should all be shifted a trifle, because of the shift of index, and therefore there's a slight shift of the energy. Although we subtract fia>/2 for the vacuum, there's a correction when we put the gas in; and this correction is proportional to the number of atoms, and can be associated with an energy for each atom. If you say, yes, but you had that energy already when you had the gas in back in the reservoir, I say, but let us only compare the difference in energy between the 25 and 2P state. When we change the excitation of the hydrogen gas from 25 to 2P then it changes its index without removing anything; and the energy difference that is needed to change the energy from 25 to the IP for all these atoms is not only the energy that you calculate with disregard of the zero point energy; but the fact is that the zero point energy is changed very slightly. And this very slight difference should be the Lamb effect. So I thought, it's a nice argument; the only question is, is it true. In the first place it's interesting, because as you well know the index differs from one by an amount which is proportional to the forward scattering for y rays of momentum k and therefore that shift in energy is essentially the sum over all momentum states of the forward scattering for y rays of momentum k. So I looked at the forward scattering and compared it with the right formula for the Lamb shift, and it was not true, of course; it's too simple an argument. But then I said, wait, I forgot something. Dirac, explained to us that there are negative energy states for the electron but that the whole sea of negative energy states is filled. And, of course, if I put the hydrogen atoms in here all those electrons in negative energy states are also ascattering off the hydrogen atoms; and theiefore their states are all shifted; and therefore the energy levels of all those are shifted a tiny bit. And therefore there's shift in the eneigy due to those. And so there must be an additional term which is the forward scattering of positrons, which is the same as scattering of negative energy electrons. Actually, for the symmetry of things it is better to take half the case where you make the positrons the holes and the other half where you make the electrons the holes; so it should be 1/2 forward scattering by electrons, 1/2 scattering by positrons and scattering by y rays — the sum of all those forward scattering amplitudes ought to equal the selfenergy of the hydrogen atom. And thats' right. And it's simple, and it's very peculiar. The reason it's peculiar is that these forward scatterings are real processes. At last I had discovered a formula I had always wanted, which is a formula for energy differences (which are defined in terms of virtual fields) in terms of actual measurable quantities, no matter how difficult the experiment may be — I mean I have to be able to scatter these things. Many times in studying the energy difference due to electricity (I suppose) between the proton and the neutron, I had hoped for a theorem which would go something like this — this energy difference between proton and neutron must be equal to the following sum of a bunch of crosssections for a number of processes, but all real physical processes, I don't care how hard they are to measure. So this is the beginning of such a formula. It's rather surprising.
716 It's not the same as the usual formula — it's equal to it but it's not the same. I have n o formulation of the laws of quantum gravidynamics; I have a proposal on how to make the calculations. When I make the proposal on how to do the closed loops, the obvious proposal does not work; it gives nonunitarity and stuff like that. So the obvious proposal is no good; it works O.K. for trees; so how am I going to define the answer for would correspond to a ring? The one I happen to have chosen is the following: I take the ring in general for any meson theory, one closed ring can be written as equivalent to a whole lot of processes each one of which is trees. I then define, as my belief as to what the ring ought to be in the grand theory, that it's going to be also equal to the corresponding physical set of trees. When I said this is equal to this. I didn't worry about gauge or anything else; what I means was, if these weren't gravitons but photons or any other neutral object — it doesn't make any difference what they are — this theorem is right. So I suppose it's right also for real gravitons, and I suppose also that what's being scattered is only transverse and is only a real free graviton with q2 = 0. Therefore, I say let this ring equal this set of trees. Every one of these terms can be completely computed — it's a tree. And it's gauge invariant; that is, if I added an extra potential on the whole thing, another outside disturbance of a type which is nothing but a coordinate transformation — in short a pure gradient wave — to the whole diagram then it comes on to all of these processes; but it makes no effect on any of them, and therefore makes no effect on the sum; and therefore I know my definition of this ring is gaugeinvariant. Second, unitarity is a property of the breaking of this diagram; the imaginary part of this equals something; if you take the imaginary part of this side, it's already broken u p , in fact, and you can prove immediately that it's the correct unitarity rule. Therefore it's going to be unitarity and so on and so on. And so I therefore define gravity with one ring in this way. Now what prevents me from doing it with two rings? The lack of a complet statement of what two rings is equal to in terms of processes; that is I can open the ring all right; but I can't put the pieces — the broken diagrams — back together again into complete sets that each one is a complete physical process. In other words some of them correspond to the scattering of a graviton, but leaving out some diagrams. But the scattering of a graviton leaving out diagrams is no longer gauge invariant, I mean, not evidently gauge invariant, and so the power of the whole thing collapses. I don't know what to do with it. So that's the situation; that's why it is crucial to the particular plan. There's always, of course, another way out. And that's the following (and that's what I tried to describe at the end of the talk — maybe I talked too fast): After all now I've defined what this results is equal to — by definition not that you should do a loop some way and get this, but that a loop is equal to this by defition, and I'm not going to do a loop any other way. But, of course, from a practical point of view or from the point of view purely of interest, the question is, can you come back' now and calculate the ring directly by some particular mathematical shenanigans, and get the same answer as you get by adding the trees. And I found the way to do that. I have another way, in other words, to do the ring integral directly. I have to subtract something from a vector particle going around the instead of a graviton to get the answer right. So I known the rule, and I know why the rule is, and I have a proof of the rule for one loop. I have two ways of extending. I can either break this two loop diagram open and get it back into the processes, like I did with the one ring — where so far I'm stuck. Or,
717 I can take the rule which I found here and try to guess the generalization for any number of rings. Also stuck. But I've only had a week, gentlemen; I've only been able to straighten out the difficulty of a single ring a week ago when I got everything cleaned u p . It's more than a week — I had to take a lot of time checking and checking; but I was only finished checking to make sure of everything for this conference. And of course you're always asking me about the thing I haven't had time to make sure about yet, and I'm sorry; I worked hard to be sure of something, and now you ask me about those things I haven't had time. I hoped that I would be able to get it. I still have a few irons to try; I'm not completely stuck— maybe. D e W i t t : Because of the interest of the tricky extra particle that you mentioned at the end, and its possible connection, perhaps, with some work of Dr B i a l y n i c k i  B i r u l a , have you got far enough on that so that you could repeat it with just a little more detail? The structure of it and what sort of an equation it satisfies, and what is its propagator? These are technical points, but they have an interest. F e y n m a n : Give me ten minutes. And let me show how the analysis of these tree diagrams, loop diagrams and all this other stuff is done mathematical way. Now I will show you that I too can write equations that nobody can understand. Before I do that I should like to say that there are a few properties that this result has that are interesting. First of all in the YangMills case there also exists a theory which violates the original idea of symmetry of the isotopic spin (from which was originally invented) by the simple assumption that the particle has a mass. That means to add to the Lagrangian a term —/Jpa^a!1 where a^ is an isotopic vector. You add this to the Lagrangian. This destroys the gauge invariance of the theory — it's just like electrodynamics with a mass, it's no longer gaugeinvariant, it's just a dirty theory. Knowing that there is no such field with zero mass people say: „let's put the mass term on". Now when you put a mass term on it is no longer gauge invariant. But then it is also no longer singular. The Lagrangian is no longer singular for the same reason that it is not invariant. And therefore everything can be solved precisely. The propagator instead of being dfiV between two currents is
V^rf
q —f*
(10)
where q is the momentum of propagating particle. The factor I/(q2—/Lc2) is typical for mass JJ, but the part —q^q^/j,2 is an important term which can be taken to be zero in electrodynamics but it is not obvious whether it can be taken to be zero in the case of YangMills theory. In fact it has been proved it cannot be taken to be zero; this propagator is used between two currents. I am using the YangMills example instead of the gravity example. I really want only the case (x2 = 0, and am asking whether I can get there by first calculating finite fi2, then taking the limit fx2 = 0. Now, with /A2 T^ 0 this is a definite propagator and there are no ambiguities at the closed rings, the closed loops. I have no freedom, I must compute this propagator. I mean there is no reason for trouble, and there is no trouble. There is no gauge invariance either. And of course I checked. I broke the rings and I computed by the broken ring theorem method a closed loop problem of fair complexity (which in fact was the interaction of two
718 electrons). I computed it by the open ring method and by the closed ring method, and of course it agreed, there is no reason that it shouldn't. It turned out that for tree diagrams you don't have to worry about this q^qv//^2 term, you can drop it — but not for the closed ring — only for tree. Therefore the tree diagrams have the definite limit as /J,2 goes to zero. And yet I have the closed ring diagram which is equal to the tree diagram when the mass is anything but zero, and therefore it ought to be true that the limit as (JL2 goes to zero of the ring is equal to the case when /j, = 0. It sounds Hke a great idea why don't you define the desired /J,2 = 0 theory that way? Answer: You can't put /j,2 equal zero in the form (10). You can't do it because of the q^qJ/J.2. So it was necessary next to see if there is a way to reexpress the ring diagrams, for the case with jx2 ^ 0, in a new form with a propagator different from (10), that didn't have a fj,2 in it, in such a form that you can take the limits as n2 goes to zero. Then that would be a new way to do the JX equal zero case; and that's the way I found the formula. I'll try to explain how to find that theory. We start with a definite theory, the YangMills theory with a mass (the reason I do that is that there's no ambiguity about what I am trying to do) and later on I take the mass to zero, then the theory works something like this. You have the Lagrangian J2(A, q>) which involves the vector potential of this field and the fields <p representing the matter with which this object is interacting for zero mass, to which, for finite mass we add the term yPA^A^. This is the Lagrangian that has to be integrated and the idea is that you integrate this over all fields A and cp; and that is the answer for the amplitude of the problem
x =
f J£^ o>^«W
DADv,
(11)
But wait, what about the initial and final conditions? You have certain particles coming in and going out. To simplify things (this is not essential) I'll just study the case that corresponds only to gravitons in and out. I'll call them gravitons and mesons even though they are vector particles. The question is first, what is the right answer if you have gravitons represented by plane waves, Ax, A2, A3 ... going in (positive frequency in Aj) or out (negative frequency). You make the following field u p . Let Azsrra be defined as a. times the wave function Ax that represents the first graviton coming in a plane wave, plus ft times A2 plus y times A3 and so on. 4 , s y m = «A1
+
pA2 + yA>...,
A+Aasym.
(12)
Then you calculate this integral (11) subject to the condition that A approaches Aa3ym at infinity. The result of this is of course a function of <x,ft,y ... and so on. Then what you want for X is just the term first order in oc, /?, y ... That means just one of each these gravitons coming in and out. That's the right formula for a regular theory, for meson theory, You calculate the integral subject to the asymptotic condition, when you imagine all these waves, but you take the first order perturbation with respect to each one of the incoming waves. You never let the same photon operate twice; a photon operating twice is not a photon, it is a classical wave. So you take the derivative of this with respect to a, /?, y and so on, then setting them all equal to zero. That's problem. (In general there's <p asymptotic too.)
719 Now the way I happened to do this is the following: Let us call A0 the A which satisfies the classical eqatiuons of motion, which i n this particular case will be
9£
9A
A"
+ {j,2A° = 0
(13)
I solve this subject to the condition that A0 equals AWTm. I n other words, I find what is the maximum or minimum — whatever it is — of the action in (11), subject to the asymptotic condition. That's t h e beginning of analysing this. The next thing is to make the simple substitution A = y 4 0 + 5 a n d p u t it back in equation (11). Then if you take jQ od AQ+B (if B is negligible you get J2 of A0 a n d so forth) so you get something like this
i[£(A,)+v>AaA„] r [£(A°+B)jO<;A°n+fi'BB+2li*AB „ „
The integral is over all B, and B must go to zero asymptotically. This business can be expanded in powers of B. £{A +B)J2(A) +^BB+2^AB = Quad (B) +Cubic (B) + ... +/u2BB. (15)
The zeroth power B is evidently zero. The first power of B is also zero because A0 minimized the original thing. So this starts out quadratic in B plus cubic in B plus etc., that's what this is here. These quadratic forms Quad (B) and so on of course depend on A 0 , the cubic form involves A0 in some complex, maybe very complicated, lockedup mess, but as far as B is concerned it is second power and higher powers. Now I would like to point something out. First — it turns out if you analyze it, that the contribution of the first factor here alone (if you had forgotten the intergal and called it one) is exactly the contribution of all trees to the problem. So that's like the classical theories related to trees. Next, if you drop the term cubic in B in the exponent completely and just integrated the result over DB, that corresponds to the contribution from one ring, or from two isolated rings, or three isolated rings, but not interlocked rings. If you start to include the cubic term is has to come in a second power to do anything, because of the evenness and oddness of function. And as soon as it comes in second power, the cubic term, having three of these things come together twice, makes a terrible thing like oo which is a double ring. So you don't get to a double ring until you bring a cubic term down to the second order. So if I disregard that and just work with this second order term Quad (B) +/LI2BB, I ' m studying the contribution from one ring. If I study this I am working from the trees. And now you see I have in my hands an expression for the contribution of a ring correct in all orders no matter how many lines come in. I also have expressions for the contributions from trees and so on. I can compare them in different mathematical circumstances, and it's on this basis that I have been able to prove everything I have been able to prove relating one ring to trees. Now, let me explain how the theorem was obtained that takes the case for the mass and for a ring. Now we have to discuss a ring, which is a formula like this X
=
y e (Quad r( B) + ^ )
m
(15)
720
The quadratic form involves A0 so the answer depends on A0 — it's some complicated functional of A0. Anyway I won't say that all the time, I'll just remember that. We have to integrate over all B. And the difficulty is — not difficulty, but the point is — that this quadratic form in B is singular, because it came from the piece of the action that has an invariance and this invariance keeps chasing us along. And there are certain transformations of B which leave this Quard B part unchanged in first order. That transformation in the YangMills theory is
$„ = £„ + ? at+ («x A) =£„ + «.„.
(16)
where the vectors are in isotopic spin space and x is considered as first order. This transformation leaves the quadratic form invariant so the Quad (B) thing by itself is singular. But it doesn't make any difference, because of the addition of the /j,2BB. If fj,2 ^ 0, there is no problem, but if fi2 > 0, I'd be in trouble. I discovered that if I make this change (16) in the actual Lagrangian and carry everything u p to second order it is exact, in fact because it's only second order. If I do it with the exact change, the thing isn't invariant, it is only invariant to first order in x. But if I make the substitution exactly, then I get a certain addition to the Lagrangian, in other words the Lagrangian of B (this includes the fj,2, the Lagrangian plus the /u2 term in B) is the Lagrangian plus the /A2 term in B plus something like this fj,2B„ • a,„ + — ju2a,„ a ; „ I have to explain that the semicolon is analogous to the semicolon in gravity. The semicolon derivative X.^ means the ordinary derivative of X minus A cross X and that's the analogue of the Christoffel symbols. Anyway, I find out what happens to L when I make this transformation. Now comes the idea, the trick, the nonsense: you start with the following thing; you, say, suppose instead of writing the original terms down, instead of writing the original Lagrangian I were to write the following:
Now I say that the integral over x is some constant or other. So all I have done is to multiply my original integral by JQ of B (by J2 of B I mean the whole thing, I mean this whole thing is going to be £ of B). If I can claim that when I integrate a I get something which is independent of B, which is not selfevident. If I integrate over all x it does not look as if it is independent of B — but after a moment's consideration you see that it is. Because if I can solve a certain equation, which is X^—JU2X = B*, I can shift the value of a by that amount, and then this term would disappear. In other words if I can solve this, and call this solution x0 and change a. to x0, then the B would cancel and it would only be «' here. I did it a little abstractly which is a little easier to explain, therefore, this term that I've added can be thought of as an integral of the following nature: Integral of some B, plus an operator acting on x (this complicated operator is the second derivative and so on) squared <Dx. And then by that substitution I've just mentioned, this becomes equal to 1/2 the operator on A
721 times x' squared Q)x, which is equal to the integral e to the one half of x times A, the operator A, times the operator A times a integrated over primed x. Now when you integrate a quadratic form, which is a quadratic with an operator like this you get one over the square root of the determinant of the operator. So this thing is one over the square root of the determinant of the operator A A. The determinant of the operator A times A is square of the determinant of A. So this is one over the determinant of the operator A, or better it is one over square root of the determinant of the operator A squared, you'll see in a minute why I like to write it in this way. In other words, when I've written this thing down I've written the answer that I want. Let's call X the unknown answer that I want. Then this is equal to X divided by this determinant's square root squared. Now comes the trick— I now make the change from B to B'. We notice that B changed to B' is simply... oh!, this is wrong, that's what's wrong, it should be just this. Now I've got it. The change from B' to B is to add something to B. Therefore to the differential of B it adds nothing, it's just shifting the B to a new value. So I make the transformation from B to B' everywhere. So then I have dx and dB, and now I have a new thing up here where I make use of the formula for J2 of B':
£{B') = £{B) + ^Bflxf,+
j
^x^x^
You see there is a certain cross term generated here and another cross term coming from expanding this out and the net result, with a little algebra here, is that becomes J2 of B, but the quadratic term doesn't cancel out and is left; there's one half of B ^ squared; that's from this term; the cross term here cancels the cross term in there; and then we have only the quadratic — I mean the x terms
And the problem is now to do this integral on x; well, another miraculous thing happens. I have the operator A, but that this down thing is xAx, and therefore its result is just determinant once; or the square of this integral is equal to this determinant, or something like that. Therefore, when you get all the factors right, X, the unknown, is equal to
X = Ue**^"*™] : [/ e^'^CDxl
Sachs: I want to ask a question about longrange hopes. Perhaps for irrational reasons people are particularly interested in those parts of the theory where is a possibility of real qualitative differences: what do the coordinates or topology mean in a quantized theory, and this kind of junk. Now I wonder if you think that this perturbation theory can eventually be jazzed up to cover also this kind of questions? F e y n m a n : The present theory is not a theory as it is incomplete. I do not give a rule on how to do all problems. I expect of course that if I spend more time on figuring out how to untangle the pretzels I shall be able to make it into such a theory. So let's suppose I did. Now you can ask the question would the completed job, assuming it exists, be of any interest to esoteric question about the quantization of gravity. Of course it would, be, because it
because of technical difficulties which are not technical difficulties just of the gravitation theory. when I am doing problems and checking. what happens to the metric. . but I won't to think about that until I have it completely formulated. so I can say very little there. So I think that you'll be frustrated b) r the difficulties that do appear whenever any theory diverges. but that's only. I don't want to start to work out the anser to something unless I know what the equation is I am trying to analyze. But at least one should certainly formulate the theory that you're trying to calculate first. there is today no expression of the quantum theory which is consistent.722 would be the expression of the quantum theory. that in such things as electrodynamics and other theories. it has not been possible to figure out the consequences of the quantum field theory in the case of strong interactions. or doing things like I just did. And so if I get my general theorem for all orders. because after all you are describing the phenomena that you would expect. in other words about the philosophy behind it. for example. before trying to do it the other way round. You say: but it's perturbation theory. and all such questions. I'll have some kind of a formulation. I worked on the thing analyzing it in the series of increasing accuracy. obviously. and if you describe the phenomena then you expect you can then find some kind of framework in which to talk to help to understand the phenomena. On other hand. But I dont' have the doubt that you will be able to do something. but exist all over the quantum field theory. there is no limit to what order of external lines are involved in the calculation of A0. and then find out what the consequences are. those I believe will be answerable. But it isn't. But even there I haven't said how many times the vector potential A0 is attacking the diagram. I do not expect that the gravitational problems will be any easier in that region than they are in any other field theory. yes. if you ask about the physical significance of the quantization of geometry. I think you would be able to figure out the physics of it afterwards. The fact is.
The formal considerations below are to give a simple explanation of the described difficulties and a quite workable recipe to circumvent them. Feynman himself described the necessary change of rules for calculation the contribution from diagrams with one closed loop. Here S[B] = / £(x)Ax is the action functional and one is to integrate over all fields B(x) with the a s 29 ._ _ _ _ F E Y N M A N DIAGRAMS FOR THE Y A N G . D. It seems that Feynman [4] was the first to show that the matter is not so simple in the cases of YangMills and gravitational fields. Bv] is invariant with respect to these transformations. A more detailed derivation of the new rules was given by De Witt [5]. It is appropriate to describe this held by means of the matricies B^x) with values in the Lie algebra of this group. The method of Feynman functional integration is used. We know from Feynman [6] that every element of the Smatrix can be written down as the functional integral (in out) ~ / exp{iS[B]} f f AB(x) x up to an (infinite) normalizing factor. FIELD 199 It is known. POPOV Mathematical Institute. and J2^ is the sum of trilinear and quartic forms in B. In this letter we propose a simple method for calculation of the contribution from arbitrary diagrams. The Lagrange function l'(. Leningrad. However it seems that nobody gave the generalization of these rules for arbitrary diagrams. The gauge group consits of the transformations where Q(x) is an arbitrary function with values in the group G. that one can associate the field of the YangMills type with an arbitrary simple group G [13].3MB„+ e[BM. N. In the quantization of the Feynman type £\ generates vertices with three and four external lines and £Q is to define the propagator function.v) = . Namely the contribution of the closed loop diagrams depends essentially on the longitudinal part of the propagator and spoils the transversality and unitarity properties of scattering amplitudes. GM„ = 3„B M . It is clear. USSR Received 1 June 1967 Feynman and De Witt showed. However the form L'Q is singular and the longitudinal part of the propagator can not be found unambiguously. FADDEEV and V. They suggested also a specific recipe for the case of one loop. that the rules must be changed for the calculation of contributions from diagrams with closed loops in the theory of gauge invariant fields. that where J2Q is a quadratic form. This ambiguity does not influence the physical results in quantum electrodynamics.I S p G ^ G ^ .M I L L S L.
24 (1963) 697. 30 . .Laplace operator in a harmonic coordinate system. but the calculation of <p[B] is more cumbersome than that of A[B]. It is appropriate to choose for the latter the "plane" c i?M = 0. This expression corresponds to the closed loop with the scalar particle propagating along it and interacting with the transverse vector particles according to the law ~ e Sp(<p[B^dll(p]). It is connected with the chosen method of extracting the fact or jfldfiCx). 96 (1954) 191. Rev./ integrate over transverse fields and A[B] is to be chosen such that the condition *[B] / r i x 5 where G(x) is a Green function of the D'Alembert operator. The diagrams appear naturally in the perturbative calculation of this integral.L.200 Volume 25B. groups and topology. x dUlG(xvx2).d^ AiAx) = const x where Q is the D'Alembert operator and u(x) are functions with values in the Lie algebra of the group GFormally A[B] is equal to the determinant of the operator Au =Uu .Yang and R.N. 4..elB^yS^u] =AQUeV(B)u After extracting the trivial infinite factor det4 Q we obtain the following expression for lnA[#] lnA[B] = ln(deU/detAo) = Sp ln(leAQlv(B)) Developing the right hand side in a power series in e we have the following expressions for the coefficients /exp{iS[Bj . 101 (1956) 1597. Phys.i i / S p O ^ ) 2 = c°nst This integral gives the perturbation series with Feynman propagator. The pure transversal Green function is to be used as a propagator for the vector particles (Landau gauge). P. 6. Acta Physica Polonica. we can say. Concrete calculations with these changes in the rules give transverse and unitary expressions for the scattering amplitudes. In fact. An other method leads to the expression dx}<p[B\ \~\AB x where the factor <f>[B] must be found from the condition <p[B] fexp{iifSp(dlx(B^)2 dx)n d n ( syu ( 5 ' J ) n / dfi = const holds. S. Relativity.. Feynman. using the natural geometrical language. 2. (Blackie and Son Ltd 1964) pp 587820. References 1.fdxnSp(Btll(x1). Ann. number 1 PHYSICS LETTERS fdx1. There exist several methods for this purpose.GellMann. We must know A[B] only for transverse fields and in this case all contribution to the last integral is given by the neighbourhood of the unit element of the group.Utiyama. 15 (1961) 437.L.Mills. 5. 80 (1950) 440.De Witt.. The idea of one of them is to integrate over the orbits and some transversal surface. R.S..BIJ) symbolizes that we x \ V.(x). It is necessary to take into account the new vertex with two scalar and one vector external line in addition to the ordinary vertices with three and four lines. In the case of gauge invariant theory it is necessary to transform this integral a little. R. It must be stressed that the Landau gauge is essential for the new rules. Then the integral reduces to the fol> lowing j exp{iS[JS]} ^B)Y\(>(djlBlJix))AB{x) X ' X ••tlxnG{xnx\) JTl^M where the factor [ [6(d. Phys. of Phys. that the integrand is constant on the "orbits" BJX~>Bjj of the gauge group in the manifold of all fields B^x). Phys. Feynman. B. There results the diagram technique with the following features: 1. Rev.B^n(xn)) 24 July 1967 x ymptote at / = XQ — ±°° prescribed by inand out> states. 3. It follows that the integral itself is proportional to the volume of this orbit which can be expressed as the integral ]\~\&Q(x) x over all matrices Q. It is the nontriviality of A[B] which distinguishes the theories of YangMills and gravitational fields from quantum electrodynamics. We conclude with the comment that one can proceed in an analogous way with gravitation theory. C. After appropriate linearization we come to the condition A[B] JYl^ip^^B^. R. Rev. This integral should be factorized before using the perturbation theory. 2.Glashow and M. P. The analog for A[B] is the determinant of the Beltrami .
so the ultraviolet problem for massless YangMills fields has been solved. which is studied in more detail.'t HOOFT Institute for Theoretical Physics. local SU(2) is broken in such a way that local U(l) remains as a symmetry.massless YangMills fields maybe renormalized. and they can be checked by algebra. which are due to an unallowed shift of integration variables in the "formal" proof. With this we mean that anomalies like those of the axial current Ward identities in nucleonnucleon interactions [2—4].:~] Nuclear Physics B35 (1971) 167188. must not occur. University of Utrecht Received 13 July 1971 Abstract: Renormalizable models are constructed in which local gauge invariance is broken spontaneously. contrary to the quantum electrodynamical case. but it is rather involved and we shall not present it here. henceforth referred to as I. In I it is proved that such anomalies are absent in diagrams with one closed loop. our prescription for the renormalization procedure is consistent. 1. . A renormalizable and unitary theory results. Weinberg [5] has pointed out that. this problem cannot merely be solved by some closer contemplation of the measuring process.1 RENORMALIZABLE LAGRANGIANS FOR MASSIVE YANGMILLS FIELDS G. with photons. owing to their large symmetry. if there are no paritychanging transformations in the local gauge group. Another model has local SU(2)®U(1) as a symmetry and may serve as a renormalizable theory for pmesons and photons. Thus. NorthHolland Publishing Company 7. It has three independent parameters.A. We do know an extension of this proof for diagrams with an arbitrary number of closed loops. it has been shown that. In one of these models. and additional neutral scalar particles. provided that a certain set of Ward identities is not violated by renormalization effects. A much more complicated problem is formed by the infrared divergencies of the system. charged massive vector particles. so we have no rigorous field theory to describe what happens. The disaster is such that the perturbation expansion breaks down in the infrared region. In such models electromagnetic massdifferences are finite and can be calculated in perturbation theory. Feynman rules and Ward identities can be found by means of a path integral method. INTRODUCTION In a preceding article [lj .
The difference is of course that we have a local invariance. In the latter case. and we have no symmetry breaking term in the Lagrangian. If. vector particles. although the Lagrangian is invariant under local gauge transformations. In sect. One statement must be made on our use of path integrals here: we only apply path integral techniques in order to get some idea of what the Feynman rules and Wardidentities might be. spin one bosons. This has been done for the models described in this paper. while the third becomes an ordinary photon. either completely. while global invariance remains evident. in which these solutions take a simple form. 2 we give a short review of the results in the preceding paper (I) on massless YangMills fields. but it surely can happen in other models. Interaction and gauge are formulated in such a way that the theory remains renormalizable. invariance under transformations of a local subgroup of the original invariance group remains evident. the physical solutions we are interested in may provide us with a certain preference gauge. and massive scalar particles. then the local gauge invariance is hidden. In sect. then they may fix the gauge. 8]. We show how the vacuum expectation value of this boson field can become nonzero due to dynamical effects. Consistency and unitarity of the renormalized expressions must always be checked later on. 3 is unitary and it is easily seen that the proof applies also to the other models. If this is the case. 4 a renormalization scheme is presented. charged or neutral. Of the original scalar fields one component survives in the form of a neutral spiniess particle. in some gauge. In sect. 3 we consider SU(2) gauge fields and an additional scalar isospin one boson. A nice feature is that in certain models the electromagnetic massdifferences are finite and can be expressed in terms of the other parameters. Our result is a large set of different models with massive. these fields have a nonzero vacuum expectation value. and the model resembles very much the massive YangMills field studied by other authors . In all these models additional scalar fields are introduced. In sect. Due to the local symmetry our models are renormalizable. but for a more elaborate description of the renormalization procedures we refer to I. and it is very well possible that all YangMills bosons become massive vector particles [6]. They all contain a small number of independent physical parameters. We do not know whether such a tiling can happen with massless YangMills fields alone. A general procedure appears to exist for deriving Feynman rules for models with a local gauge invariance. 5 we prove that the model of sect. causal. 6 we describe an example where local invariance seems broken. or partly. All YangMills particles get equal mass. photons. of which we present some.168 G. 't Hooft. The transition from a "symmetric" to a "nonsymmetric" representation is done in a way analogous to the treatment of the omodel by Lee and Gervais [7. and how two of the YangMills bosons become massive. and some of the YangMills bosons remain massless. and unitary. In sect. oppositely charged. which are representations of the local gauge group. Massive YangMills fields However.
The model can be used to describe pmesons as elementary particles. and it may be chosen in an arbitrary way. The Lagrangian is £ V M = . 't Hooft. only depending on the divergence of the field W°. because of the gauge noninvariance of J3C.G.1) ^ 5 3 / + 4 ^ C ' (24) Furthermore.\Ga Ga +J2c(d W). 7 it is shown that our "pmeson model" can be enriched with electromagnetic interactions without destroying renormalizability or unitarity. From (2.1) the Feynman rules may be constructed by ordinary Feynman path integral methods: the procedure is clarified in the appendix. where Du is the covariant derivative. because they cancel the helicity0 states in the Wfield. with Ga = 9 Wad Wa+gf.3) (2.1): Under an infinitesimal gauge transformation generated by Aa(x). the quantity 9 Wa . The ifi particles (and antiparticles) do not occur in the intermediate states in the unitarity condition. J3 C is an extra term. defined as (2. 10] except for the presence of one extra neutral scalar boson with arbitrary mass.3) is related to the behaviour under local gauge transformations of the gauge noninvariant part J? c of the Lagrangian (2. But.2) v ixv n v v M abc M v ' a where W are the YangMills field components and/ a ^ c are the structure constants of the underlying gauge group. an extra factor — 1 must be inserted in the amplitude for each closed loop of I/J'S. p° — y mixture leads to phenomena like vectordominance. an extra complex ghost particle field ^° must be introduced. Massive YangMills fields 169 [9. described by the Lagrangian J2^dy\D^f. The Lagrangian (2. WbWc. (2. In sect. g is a coupling constant. thus fixing the gauge. RESUME MASSLESS YANGMILLS FIELDS The massless YangMills field has been discussed in I. In the appendix we formulate the Feynman rules for the various models. 2.
These Ward identities supply a unique prescription for the subtraction constants in a renormalization procedure. The choice £c=_a(d^wa_ca)2. tiius ensuring local gauge invariance and unitarity.5). which is precisely the i^particle contribution. Massive YangMills fields transforms as: *X=d*KS~ldWf (25) Because the fields Wa occur explicitly in the covariant derivative Z) in eq. SELECTION OF A PREFERENCE GAUGE BY INTRODUCING AN ISOSPIN 1 SCALAR FIELD Consider the case that the local gauge group is SU(2). t Hooft. a ^ x (26) leads to the Landau gauge for C = 0.3) from eq. and suppose that an isospin one scalar field or current Xa{x) exists which has (in a certain gauge) a nonzero vacuum expectation value [6]. 3. and from the fact that the amplitudes must be independent of Ca(x). If we introduce other fields which are representations of the gauge group.7) k2ie (2. and from them unitarity of the system follows.8) and again one can derive Ward identities.1) for all x. (2. in which the propagators are (2. For / = 0 we have the Feynman gauge. (2. which may be taken to be the gauge in which X1{x) = X2(x) = Q (3. a nontrivial Jacobian factor is needed in the gauge dependent expressions for the amplitude. The most appropriate choice however is £c=K\Wfc)Ja(x))2. In appendix A of I it is shown how to derive the Lagrangian (2.170 G. then all derivatives in their Lagrangian parts must be replaced by covariant derivatives. one can derive Ward identities. However. Then this isovector is apt to select a preference gauge.5). a gauge transformation of the original system to this "Jffield .
then this energy can be written as an integral over space pf an Hamiltonian density 9((x). In the symmetric representation the Lagrangian is £ = £ Y M Kzy02*M2X2£A(X2)2 . but its formal possibility indicates clearly that the components X1 and X2 are unobservable. In this approximation all fields may be considered as being classical. In order to get some insight in what might happen let us consider the treeapproximation. we disregard all graphs with closed loops. Thus we have three independent parameters g.1). (If we specify the gauge. By a global gauge transformation this vector ea can always be pointed in the zdirection.G. (3. where all fields are constant and the total energy has a minimum. in general we shall abandon the gauge (3.2) This Lagrangian corresponds to a renormalizable theory. 't Hooft.3) must be replaced by '°\ <0X (x)0> = F\ 0 1/ (3. and W ~ 0. In the latter case we expect that the A'field is nonzero in the equilibrium state: for slowly varying Xa. (3. The last term is necessary for fixing the counterterm in divergent graphs with four Xlines. Massive YangMills fields 171 gauge" would in general involve a nonpolynomial Jacobian factor (cf. I). but /U2 may have a negative value. (3. and we fix the gauge by adding some functional J2C to the Lagrangian as in sect. we have 9t~ J dx(WX2 + k\(X2)2). Eq. that is. In our renormalizable model.4) a '¥•+<**): (the parameter X is of order g2). thus destroying renormalizability of our model.) In order for this vacuum to exist. X is simply a boson field. 2. So. the parameter X must be positive. v This has a minimum for X"(x) = e"J^^. and the vacuum corresponds to the equilibrium state. . /i and X. and X^ acts as a "schizon": our symmetry seems to be broken. If we do quantum mechanics things do not change drastically.3) with ea an arbitrary vector with modulus unity.
One could very well proceed like in sect.8) In here the functions Ja(x) will be put equal to zero. We deliberately have not yet specified the local gauge. The fields in eq. (3.\W2A2 . We write t X\x) with <0\Aa(x)\0) = 0. We have seen that the gauge (3. but the resulting Feynman rules are rather complicated and a massless ghost remains. but the fact that all physical amplitudes are independent of them enables us to formulate Wardidentities.{(D^A? . in accordance with eq.7) Note. (3. Massive YangMills fields We now proceed as Lee did for the amodel [7].8) transform as follows under infinitesimal local gauge trans .\WA2A3\\(A 2 2 ) g2FA3{W^ + wf) + g2FW*(A] Wl + A2 W2J . The Lagrangian (3. (3.hg2F\wl2 + Wf) . that the "tadpole condition" (3.172 G. where /3 = fi2 + i \ F 2 . 2 and choose the local gauge by adding J2C=£CR> W)=h{b WaJ(x))2.6) .6) implies that /3 = 0 in first order of g and X.4).1) is no good. t Hooft.2) then becomes £ = £ w = F\Q l + Aa(x).0 ( M 2 + FA3) + gF(A]dilwZA2bllW1J. because it renders the theory unrenormalizable.5) (3. (3. (3. It is more convenient to choose: £c=h$XJ3(x)?\(*yiigFA2J1{x))2 Hd^+gFAjJ^x))2.
* / M 2 g~13M(Z?MA)7 + gF2A! .H) and#. Massive YangMills fields 173 formations: 3 W3'=b W3g~lb M M M M ° (D A)3 . (3. the i^ghost restores unitarity. There are two different types of ghosts: 2 M (3. Mv M ' (3 M ^ .G.« E 4 2 ) ' = 9M ^ . with the propagator 5 UK k .te and a neutral.12) a massless photon W . Because of our choice (3. with mass M\/a. (3. . a= \/g2>0. The resulting theory has two massive. as is shown in the appendix. 't Hooft. with the propagator 5 MV k +M2ie 2 M ' (3.9) 3M(£M A) 2 + gF 2 A 2 + gFelbc AbA£ .13) % Note that this expression is not Hermitian. 2 we derive the ipghost LagrangianJ: J2ip=i>y\D^yg1F\**1v1 +/V) (3.8) for £c the last term in (3.7) is cancelled. With the same procedure as in sect. scalar particle A3.10) ^2F(^e2bc^Ac/2e]byAc). The heuristic Feynman rules are here the correct ones.gFe2bcAbAc (3 M H/2 + ^ 4 7 ) ' = 3„ Wj + gFA}g1 . charged vector particles W ± iW . Because of these features Feynman rules must be derived by path integral methods. likewise the term + 4(3 W)2 in \Ga Ga . Let us finally replace the three parameters by M = gF>0 .
with no minus sign. However.8) for J3C.6) holds.3 / ( f l / .13). and the A* particles. from the Lagrangian 2 = \Ga Ga . get lost. in sect. RENORMALIZATION The expression (3. The Feynman rules for the contribution of their Fourier transform JJJc) are given in . quadratically convergent propagators (3.174 G. (3. and massM2 (resulting from a contribution of J2C.8)). with positive definite metric. t Hooft. and all other Feynman rules. and therefore also unitarity.15) The functions Ja(x) are arbitrary. They may be chosen to be zero.i ( 3 Wa)2+ £ \{D A)2\M2(Wl2+W22) + gMW*(Aj M>1 +A2W$X [b^MAz] with (S(U2 + jA3) + J3(x) 9 M w£ + / y (x) (3. except for the above mentioned minus sign.12) and (3. Second: the real A ^2 ghosts. may be derived with ordinary Feynman pathintegral techniques. 5 we derive that these particles cancel the unphysical polarisation directions of the R'fields in the intermediate states. A minus sign must be inserted for each closed loop of vsghosts. that the tadpole condition (3. 4. The most compelling reason for this is the unitarity condition.12) and (3.13). has especially been chosen in order to acquire the simple. resulting from the anomalous propagators (3. Thus.3 f V l 2 + ^\2)+sM(^1e2bc^2elbc)<pbAc . Massive YangMills fields First: the complex i^ghost with the oriented propagators: (k2 + M2 — ie)~~ * for ip1'2 and (k2 — ie)~ l for tp3. These. To this purpose we can use the observation that in all orders of g the physical amplitudes must be independent of the source function Ja{x).14) +J2(x) [drf+MAj] \J2a{x) £v= . The constant j3 must be adjusted in such a way. The complex ipparticles with their unphysical "Fermi statistics". and in general this modifies the theory such that the symmetry. must all be considered as ghosts. (3. eq. and to arrive at Feynman rules which are those of a renormalizable theory. the cutoff procedure must be chosen in accordance with our symmetry requirement. which enables us to formulate Ward identities. renormalization requires the introduction of a cutoff procedure.
3).14)). a variation upon these identities can be found for the case that one of the external Wlines on mass shell has a nonphysical polarization direction (fig. also in this model. (Note that (d) is cancelled by contributions of bare W and . . 't Hooft.) A graphical notation for the Ward identities is shown in fig. 3 can be proven either by combinatorics.3 '. 1 (compare eq.2. or by using a formulation in terms of path integrals: one must consider an infinitesi (off mass shell) ln> Fig.2. fig.5 ab (d) Fig. Further. Example of Waid identity expressing the fact that physical amplitudes are independent of the /source.4 particles. 2. A combinatorial proof of these Ward identities can be given in the same way as in the case of the massless YangMills fields. The number of "/lines" must be nonzero. Massive YangMills fields 1. 2. 1. The identity in fig.3 175 X— ik (a)  *2 (b) (c) <k> x = x JbCK) . (3. Feynman rules for the contribution of the source function Ja to the amplitude.G.
a unitary field theory of massive vector particles would have propagators . 5. Thus the number of subtractions has been determined. where (k) stands for the momenta of the outgoing lines. and £ is a logarithmic factor. 2 and 3 may be used to prove unitarity of the model. The procedure sketched here can be proven to be consistent as soon as some gauge covariant set of regulators is found. whereas the Ward identities give a large restriction on the possible values of the subtraction constants. Such a set can indeed be formulated for diagrams with one closed loop by the introduction of a fifth dimension in Minkowsky space (cf. is to apply subtracted dispersion relations.1) (J(x) being also infinitesimal). but we shall not present it here.176 G. One of the simplest ways to use these Ward identities to calculate the renormalized higher order corrections to the amplitudes. Ward identity for the case that one of the external H'lines on mass shell has a nonphysical polarization direction. Massive YangMills fields Fig. 't Hooft. Actually. The proof that the contributions of the y? and ip? intermediate states cancel those of the unphysical W states is the same as in I and will not be repeated here. mal gauge transformation generated by Aa(x) with the property diiDtxA(x) = J(x). The procedure is analogous to the proof given in (I). By introducing more dimensions one can give a consistency proof for all orders. (4. As to the charged particle states. it is enough to show that the residues of the poles at k2 . I).0 of these propagators cancel. the amplitude for a diagram with N outgoing boson lines must behave at infinity like (k)^~N L(k). The behaviour of the amplitudes for the momenta going to infinity is prescribed by the condition that renormalizability must not be destroyed at higher orders: hence. 3. UNITARITY The equations shown in figs.
12). A2 and i/>particles. anomalous k k UU l^(k + M 2 . We like to mention in more detail a very interesting case: local SU(2)gauge invariance with an isospin2 "symmetry breaking" field K{. In fig. indeed. 2 and 3. 4. the W12 propagators have anomalous parts i^^v 177 M (k +M2ie) 2 2 (5. Hence. we see that as a consequence of the equations in figs. as unphysical. Many different models may be constructed this way. and the Aj. charged. They can be left out of the intermediate states without invalidating the unitarity equation. 6. vector particles. (Note the explicitly written minus sign for the i^loops. It is because of this phenomenon that we can consider the anomalous part of the Wfield. Massive YangMills fields instead of (3.2) Now. unitary theories with massive. 't Hooft. given in (I). Combination of unphysical propagators for fixed value of k.i e ) 2 anomalous £> 4 />~\ > k <P'~ \ k Fig. 4 it is shown which combination of these propagators has to be considered in order to prove this cancellation. ISOSPIN4 SCALAR FIELD Our most important conclusion from the foregoing is that a basic principle like gauge invariance can lead to renormalizable.G.) For more details of this proof we refer to the treatment of the analogous case in massless YangMills fields. ) = * I • \K'2 + iKp . the integration over k has not yet been carried out. the residues of the poles at k2 = — M2 of the unphysical propagators cancel.
4) \Q' V 2 \_yh +jib .t Hooft.n2K*K . In this representation. It is of order g2. (6.2) For negative values of /j2./ and to introduce the independent parameters M2 = \g2F2 . andg.178 G.3) Now. (6.5) a = \/g2 . the Lagrangian is: £ = \G«vG°v + + £c\M2 W2 i(3 M Z) 2 \{b^)D^ f27 $4aM2Z2 hgWl{Zd^a^b^Z)\g2wlw2+Z2)\gMW2ZclMgZUj2+Z2) iag2W2+Z2)2P kiz2 + ^2) + fz Mil a9 MWaM . . it appears to be convenient to express the complex spinor K in terms of a real isospin singlet and a real triplet: (6. the field K is expected to have a nonzero vacuum expectation value.(D^K)' D^K . Massive YangMills fields Let the Lagrangian in the symmetric representation be £ =£ YM . T (6.\\{K*K)2 (6.2 + \F2 must be chosen in such a way that all tadpoles cancel.1) The covariant derivative of an isospin! field is: DllK = dllKiigi"WlK. which by a suitable global rotation in isospin space can be taken to be: (D\K{x)\Q) = F 0 (6.6) where the parameter 0 = y.
Only local gauge invariance has been broken. and all three isospin components of the ^fields have become massive.G. Many authors [9. For large values of this mass we get something very similar to the old model (6.6) as well as the Ward identities remain invariant under global isospin transformations. But now we see that the introduction of one physical isospinzero particle Z can render the model renormalizable.9). Feynman rules for the /source contribution to the amplitude. j a (k) W* x * ik J (k) a x ja(k> * = x vk) . if the \p fields are considered as a triplet and the Z as a singlet. (6. . By studying the behaviour of £c under gauge transformations we derive the Lagrangian for the ghost field <p (compare sects. 12]. In our model also two ghosts appear: the complex ipfield.8) The Feynman rules for the source function contributions are now those of fig. and the i//field. Massive YangMills fields £c is chosen to be 179 so that again the Feynman propagator (3. Here the i// fields act as additional ghosts.12) for the PVfield emerges. 't Hooft. 5. although many of the divergencies can be seen to cancel by the introduction of two kinds of ghost fields [11. The Z is an additional physical particle.6) cancels. 2 and 3) £^= b^*D^M2<p\\Mg*\Z +\MgeabcS%br . Now we observe that the Lagrangian (6. Its mass is a new independent parameter. and the last term of eq. (6. 10] have considered the massive YangMills theory described by the Lagrangian That model appears to be non renormalizable. 5. with Fermi statistics.Bab Fig.
Let in (6. ISOSPIN AND ELECTROMAGNETISM. His lepton model can be shown to be renormalizable.4) f The model of this section is due to Weinberg [ 13]. Let us consider an infinitesimal gauge transformation: /r' = ( l . W'^W'g^Af. = A2 = 0 . electromagnetism can be introduced in an elegant wayf. VECTOR DOMINANCE In the previous model.i i A a 7 * . (7.2) Al = A^q\A. that are the transformations with A . (7. Consider first the symmetric representation (6. 1 . where Aa(x). who showed that it can describe weak interactions between leptons. Now if the K field has a nonzero vacuum expectation value: <0*(JC)0> = F ( ) . A J = . A(x) are generators of an infinitesimal gauge transformation. Massive YangMills fields 7.1). The Lagrangian is then: 2 = £YM tfjF^(D^K)*D^Kti2K*K±\(K*K)2 with . which does not break isospin. (7.3) then the physical fields will only appear to be invariant under those transformations (7.2 A = AEM.180 G. Let us assume the presence of a "hyperelectromagnetic" field.i A K . Au.2) that leave the spinor 0 invariant. t Hooft.1) DpK&D^K + iqA^K. (7. The gauge group in this model is SU(2)®U(1).1) only the K particle have a "hypercharge" q.
^ M p 2 z .8) . (.G.k 2 P 2 ( ^ 2 + z2) EM M*jPJ aMgZty2 +Z 2 )W 2 (* 2 +Z2)23 (i(Z2 + <p2) + —z) (7..5) VQ/ V2 \. Finally. we make the substitutions: *. (i •£)"*.7) In terms of these variables the Lagrangian (7.2.6) P3' = P3 p ^ / p ^ e ..•£)'.^ t ^ ) .1) becomes: ^ M ^)^ M ^ + k p ^ ( Z 3 f ^ a .^ a M z ) . . Massive YangMills fields 181 Let us call such transformations electromagnetic gauge transformations. If we define Wl2^p1'2 W3 = p3 + ^A (a) (b) (7. 't Hooft. (7. (7.^2 + i^]t then these quantities transform under electromagnetic gauge transformations like: Z' = Z ..
and the number of independent parameters is small.9) ft c abe nHv Let us choose £c= i(3f p^^a7fl)2_i(3M^/EM)2 .(3M ± ieAJWj ± W2) .ii) again with the additional minus sign for each closed loop of </>'s. (8.12) Note the third term in (7.12) and (3. and the pparticle have the Feynman propagators (3. are not needed for renormalization. that there are no ambiguities due to infinities.10) so that the last term in (7.5b).182 G. (7. So.^ ^ . M r 3 n 3 ' 3 j f V . (7. 8.8). in the asymmetric representation. For instance. It is a consequence of the translation (7. CALCULATION OF ELECTROMAGNETIC MASS DIFFERENCES One of the main virtues of the models presented here is. M.. which leads to phenomena like vector dominance.13) resp. a and e. we have arrived at a renormalizable model containing photons. in the model of sect. pmesons and neutral Zparticles. The contributions of the i/5ghosts is described by V . .. •f It appears that counterterms of the form NaeabcXbNc. a s3EM ^M" M a_3EM « * x «+ 'M (7. There are four independent parameters: g. Naea/jCA^Nc.a ^ X ^ . 3 one may introduce an isospin one fermionf J2N=N(m+yllDJN.8) is cancelled. The parameter (3 is dictated by the tadpole condition: <0Z(x)0> = 0 . t Hooft. while the photon. It is interesting to calculate some "electromagnetic" mass differences. Massive YangMills fields where a E M iK = a 4.^ ' ^ + ^ v ' V ^ . (7.1) or. ± i^2) .
't Hooft. Massive YangMills fields 183 A direct computation of the difference in mass of the A'1 and A*' seems to lead to ambiguous results because the integrals for the selfenergy corrections diverge. Let us consider all terms of this graph up to third order in g (fig. 6b will not contribute in the third order of g. The first term only contributes if m± ¥= m°. 3. . and the neutral Nj are both on mass shell. Hence the last four diagrams of fig. But. augmented with isospinone fermions. 6b). so q can be taken of order g 2 . All contributions up to order g3. a. Ward identity for the model of sect. J (off mass shell) = 0 (a) (0) = 0 Fig. if the charged particle N2. b. one of the Ward identities states that the graph of fig. 6a equals zero.G. 6. Now the mass difference m± — m° will be of order g^.
B=0 (8. The integral in B converges.. and the result is: m° 2g2 (4TT)2 r . large values ofM2/m2 eq. 6b. (8. The momentum transfer q may now be taken to be zero.4) 2 2 {M >m ). M. The term which breakes the gauge invariance is always called £c. The same procedure has been applied here. and both external Nlines have momentum p with p2 = — m2.2) where B is the second graph of fig. As a consequence of this procedure. (7. Massive YangMills fields Thus. 7. (8.Veltman for his invaluable criticism and encouragement. .3) may be simplified to m±~m° = ^ 87T (M2 < m2). It was shown how to make the integrand gauge noninvariant without changing the total amplitude. the Feynman rules must always be derived . This massdifference is of zeroth order. This massdifference is always positive. t Hooft.184 G. and originates from the "vector dominance" term in eq. and for small resp.3) (in second order of g). Diagonalisation of the bilinear terms in (7. g .8). w±_mo=3^mlog^i {An)2 m2 A negative massdifference is found for the pmeson itself in the model of sect. H M2{\ x) L rrihc2 • (8.. the Ward identity reads g(m±mo)M.8) leads to the mass formula for the pmesons (in zeroth order): M 2 = _^!i__ 0 (85) \e2lg2 The author wishes to thank Prof. APPENDIX Feynman rules for the various models In the preceding article I a system with a complete local gauge invariance was quantized using a path integral technique.
VW3 li V nv k +M22 5 5 ie (A.G.ie 2 (A. Model of section 3. and the vertices are the coefficients in front of the remaining terms in the Lagrangian.4) ^. physical particles: Wl. regardless whether time derivatives occur or not.1) M» V k ie 1 k2 + aM2 . This means that the propagators are always the inverse of the matrices in the bilinear terms of the Lagrangian.3) ghosts: A l.2 2 2 1 k +M ie (A. I3 A agb (A.2) A3 (A.7) for each closedI loop of ip lines: W X^ *<p as in massless YangMills fields.6) 1 (A. some of the vertices: 1 k . We mention here some of these Feynman rules.8) u''' ^ f2 A . Massive YangMills fields 185 from the final Lagrangians by performing the path integral.14) and (3.5) _y—.15)). and not by canonical quantization.2 i[compare (3. 't Hooft.2 2 2 1 k +M ie (A.ie 2 (A.
3agM A3" ^A3 (A. Massive YangMills fields .8)).ie (A. 11) yjft'^ A3 g (A. 12) A iJ_ l. 10) ^Cifc: (A. with /3 chosen such. Ac 1 J>^ __ J_ • Ac  v v (* = 0) A S^26C (A.6). 14) (A.7) and (6. 15) ghosts: ^i_ 1 k2+M2ie (A. 't Hooft. 12) and (A. 5 Hi> k +M2ie 1 k2 + 4aM2 . 16) . (6. that all tadpole contributions cancel.13) The vertices (A.186 A3 G.2 • — « A3 A3 1 0 (A. physical particles: W M Z V 2 (compare (6.13) must be added to higher order tadpole.and selfenergy diagrams.9) etc. Model of section 6.
(compare (7.11)).8). Rules as in preceding model. but with additional photon lines: A At " 5 M» A: .24) etc. Model of section 7.25 > .9).21) vertices: A ™^~ A V1 P ^M p3 e{kh g tiv M (A. e5w ( A .C. (7.ie 2 (A. (7.23) 2 «>" " """•*.20) etc.» e(p<7) u (A.q). Massive YangMills fields some of the vertices: Wb >?N \pa'p q'^c Z X 187 kigealJPl\ (A18) 2<*Mg8ab (A. (A. 't Hooft. 19) z 1^ 6aMg (A. The /source can now emit a photon and a pmeson simultaneously: Jit^X Pu etc.10) and (7.22) < T V 1 eip .
Nucl.Gerstein and RJackiw.22) as it stands. Nuovo Cimento 60A (1969) 47.25) for the /source contribution may be avoided by choosing another expression for £c (eq. B.G.L. of Phys.W.Bell and R. Phys. Nucl. J. Phys. M.Esposito and S. Phys. Ann. Phys. Phys. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] G.Termini. Letters 19 (1967) 1264. B33 (1971)173. Rev. U. I. S. Rev. Moscow preprint 1970.D.S. 140 (1965) B516.Fradkin. Rivista del Nuovo Cimento 2 (1970) 498. B12 (1969) 627.Weinberg. B9 (1969) 649. 155 (1967) 1554. t Hooft. If necessary.8) completely. Phys. J. Phys. Rev.L. Nucl. (7. L. vertices like (A. Phys. 56 (1970) 140.Kibble. Massive YangMills fields The vertex (A.W.Gervais and B. Nucl.B. For small e it is easier to leave (A. Phys.22) is a consequence of the fact that we did not diagonalize the bilinear terms in (7. S. Phys.A.188 G. T. [ 12] E.10)).Lee.W. Rev.Lee.Weinberg.Jackiw. D. [13] S.'t Hooft.S. A. B7 (1968) 637.Boulware.Faddeyev. The diagonalized propagators have a rather complicated form. . Rev.Veltman. B21 (1970) 288. 177 (1969) 2426. 181 (1969) 1955.Adler. Massless and massive YangMills field.Slavnow. Nucl.S.
This property resides in the fact that zeropoint field oscillations increase the effective charge not in the highmomentum region as in QED [7]. Leningrad 188350. A subsequent analysis of perturbation theory in such theories (Politzer [4]. This gave hope that such theories may incorporate the phenomenon of color confinement which is fundamental to present day ideas concerning the structure of hadrons. Khriplovich [6]) has shown that they possess a remarkable property called asymptotic freedom. GR1BOV Leningrad Nuclear Physics Institute. i. it is very likely that this improvement reduces simply . the problem of color confinement is closely connected with that of the quantization of large nonlinear oscillations. USSR Received 13 January 1978 It is shown that the fixing of the divergence of the potential in nonAbelian theories does not fix its gauge. when integrating over the fields in the functional integral. This limitation on the integration range over the potentials cancels the infrared singularity of perturbation theory and results in a linear increase of the charge interaction at large distances. Introduction The quantization problem for nonAbelian gauge theories within the framework of perturbation theory was solved by Feynman [1]. Thus. Answering the question as to whether color confinement occurs in nonAbelian theories proved to be a very difficult problem since the nonAbelian fields possessing charges ("color") strongly interact in the largewavelength region. Gross and Wilczek [5]. The ambiguity in the definition of the potential leads to the fact that. but in the lowmomentum region.e. DeWitt [2] and Faddeev and Popov [3].N. it is apparently enough for us to restrict ourselves to the potentials for which the FaddeevPopov determinants are positive.Nuclear Physics Bl 39 (1978) 119 © NorthHolland Publishing Company QUANTIZATION OF NONABELIAN GAUGE THEORIES V. at large distances between the charges. 1. As will be demonstrated. for which the substantially nonlinear character of nonAbelian theories is decisive. In this paper we show that in the region of large field amplitudes the method of quantization by Faddeev and Popov is to be improved. Gatchina. Strong interaction between vacuum fluctuations in the region of large wavelengths means that at these wavelengths a significant role is played by field oscillations with large amplitudes.
which consists in integrating only over the fields for which the FaddeevPopov determinant is positive. but goes to infinity at infinitely large distances between charges. llcU. This in turn results in the fact that the "effective" charge interaction does not tend to infinity at finite distances as occurs in perturbation theory. (4) in Euclidean spacetime and imagine the functional space Au in the form shown in Pyb^'h "^Air* ^v~y^' Fig. 1 . GribovI Quantization of nonAbelian gauge theories to an additional limitation on the integration range in the functional space of nonAbelian fields. A conventional method of relativistic invariant quantization [3] is as follows.f1A. but substantially reduces the effective oscillation amplitudes in the lowfrequency region. S+ = S\ (3) contains nonphysical variables which must be eliminated before quantization. . if at all. This additional limitation is not significant for highfrequency oscillations. 2.N. being invariant with respect to the transformation All = S+A'llS+S+dllS.v dfdAu d^Ay + \A. Nonuniqueness of gauge conditions The difficulties in the quantization of gauge fields are caused by the fact that the gauge field Lagrangian £= 4pSpi>7> 1 (1) (2) r fj. Let us consider a functional integral W = Je r J£d\ r.p J where A^ are antihermitian matrices. Sp A^ = 0.2 V.
[V M 04')Sa M S + ]}S]. (11) This conclusion is correct under the essential condition that. one should integrate over matrices S and fields A^ which have a certain divergence / = 3M/1M. a situation of the type where many . i. Gribov / Quantization of nonAbelian gauge theories 3 fig. However. instead of integrating over A'u. eq.N. with the integration over S omitted. The Faddeev and Popov idea is that. Then W is written in the form J J A(A) [VniA^Sd^S^jS]. (6) Since the variation with respect to S of the expression under the sign of the 6 function is dM [VM(A). (7) where the operator 13(A) is defined by the equation 3c4)V/ = a2i// + 9M[/iM^] = a M [ v M ^ ) ^ ] . (5) xfdSS+8[fS+{dnAl+ where A(A) = fdSS+8[f~S+{btiA'n+ vMC4') = a M + 4 . (3) defines the line L (as a function of S) on which £ i s constant. one can always find a unique field A^ with a prescribed divergence/. where one cannot find a field A^. respectively. (10) Since (10) does not depend upon/. we may integrate over /with any weight function. given a field A'^.e. In so doing. Replacing in (4) the variables (8) we obtain W = Je *£i x b(fdtlA^)dA  5(4)11 dS • S+ . there are neither situations where curve (3) crosses the line 3M/1M =/several times (curve L') nor where it does not cross it at all (curve L").V. where the transverse and longitudinal components of the field A^ are plotted along the horizontal and vertical axes. We do not know any examples of situations of the type L". S+ dS] then 1 A(A') r=ll504)ll. W is obtained in the form W = fexdf£d*x +~ S p / O ^ ) 2 d4x\ SC4) <L4 . 1. Then for fixed A^. which is gaugeequivalent to a given field A'u. with a certain divergence. exp {(\/2<xg2) Sp J f2dx} being commonly used for this purpose.
N. (13) has solutions for S which tend to unity at r > °° and result in rapidly decreasing A2ll. (13) reduces to the Laplace equation 3 ^ = 0. (13) for . of the transverse potentials equivalent to the vacuum. though corresponding to such S that do not tend to unity at infinity. . (13) tending to unity under r *• °° which depend on the character and the magnitude of the field A^. where S= el{f is a unit matrix. but we shall demonstrate below that there exists a possibility of a sufficiently universal solution leading to physically interesting results. i. or obtained from it through the substitution ofA2fl for A lM and 5 for S. (14) + (12) (13) and to eliminate nonuniqueness it is sufficient for us to confine ourselves to the fields which vanish at infinity.4 lM = 0. we eliminate the fields gaugeequivalent to "small" fields. In a nonAbelian case. eq.Under these circumstances. In the appendix we shall also show that. For instance. are in order of magnitude similar to a set of solutions to the Laplace equation. This problem seems to be almost hopeless. is possible. but that all of these. there should be a unitary matrix S connecting A lM andv42M. (11) by the expression where N is the number of fields gaugeequivalent to a given field A and having the same divergence. and satisfying the equation 9 M 5 + [V M C4 1 ). when integrating only over^4M vanishing at infinity faster than \jr. but for large enough A^ the gaugeequivalent fields will remain and hence N(A) in (15) will be needed. when both things are required. result in the potentials A2il decreasing as \jr. Gribov / Quantization ofnonA belian gauge theories gaugeequivalent fields A^ with a given divergence correspond to a given field A'^ is typical in nonAbelian theories.4 V. the problem of calculating N(A) reduces to the analysis of solutions of eq. from which it will be evident that a set of these solutions. In an Abelian theory. the nonlinear equation (13) cannot have growing solutions and hence even for A lfl = 0 it has solutions for S leading to a decreasing A 2M • In the appendix we consider examples of the solution to eq. in order for two gaugeequivalent fields Alfi andA2fJi with the same divergence to exist.^]=0. or restrict the integration range in the functional space so as to have no repetitions. to calculate correctly the functional integral in a nonAbelian theory. we must either replace eq. In this case. A7p = S+A1^S + S+dfxS. An intermediate case. Indeed. with values of/l l M large enough.e. which grow at infinity.
but also for any field in the region C„ there is an equivalent field in the region C„_].. the equation 504)<// = e\p (18) is solvable for positive e only. there exist a given number of bound states for the FaddeevPopov ghosts. are located on different sides of this line. 1. they will always lie on different sides of the corresponding curve C2. •••» £« o n which the ghosts have zeromass levels. Fig. one can imagine the fields for which eq./.V. (17) a+ = a. Clearly. it may be asked whether two near equivalent fields that can exist close to the line.e.. 2 shows this division of the field space into the regions C Q . we draw the conclusion that the field A^ can only have a close equivalent field when the FaddeevPopov determinant for this field turns into zero. Gribov / Quantization of nonAbelian gauge theories 5 3.e. if there are only two near equivalent fields. (17) is solvable as dividing the functional space into regions over each of which eq. or may be arbitrarily situated.field. let us see for what fields A^ there exist gaugeequivalent fields close to the former and having the same divergence. etc. i. a] = 0 . 8. another in Cj. Hence. say.e... (18) has a given number of eigenvalues.N. C„. and eq.e. i. Substituting into (13) S= 1 + a. (13) with S close to unity.. Moreover. then there are no bound states in such a.. 2.e. Thus. C J . bound states.. A limitation on the integration range in the functional space In order to gain some insight into the nature of nonuniqueness in the functional space v4M. i. over which the ghosts have 0. we would prove that instead of integrating over the entire space of the fields A^. if we imagine the space of the fields A^ in this way.1 there is an equivalent field within the region C0 close to the same curve. (17) is simply an equation for the eigenfunction of this operator with a zero eigenvalue. we get S04)a = 9M[VMG4). if the field A^ is sufficiently small in the sense that the product of the width of the region where A^ differs from zero with its amplitude over the same region is small. If we could prove that not only for small neighbourhoods close to the curves £„. i. For a particular still greater magnitude of the field. indeed. which becomes one with a negative e as the field increases further. i. it would be sufficient for us to confine ourselves to the region C 0 . We shall demonstrate below that. the level with a zero e reappears. (16) Since D(A) is the operator whose determinant enters into the functional integral. by the lines Cj. . For a sufficiently large magnitude and a particular sign of the field there appears a solution with e = 0. fi2. . or (which is the same) if the field is such that the FaddeevPopov ghost has a zeromuss bound state. one field within the region C 0 . we shall show that for any field located within the region C{ close to the curve S. to integrating only with respect to the fields Au which create no bound states for the ghosts (up . what are the conditions for solving eq.
*o]=0. i. Proof of the field equivalence over the regions C0 and Cj close to their boundary We shall first of all show that if the field A^ is close to S. We write the field Au in the form ^M = C M +a M > (19) where CM lies on £j.N. and aM is small compared to CM in the sense defined below. Gribov / Quantization of nonAbelian gauge theories Fig. they are separated by a finite region from the boundary fij and are in no way connected with the region of small fields A^ lying within C0 for which perturbation theory holds. (13) always has a solution with S little different from unity and tending to unity at r * °°. We shall assume below (until the contrary is established) that these additions are either nonexistent or insignificant and that the integral (15) is determined over the region C 0 . (20) . i. 2 to the first zero of the FaddeevPopov determinant).. However. 4.e. we must retain A%4) in (15) because we have not proved that there are no equivalent fields over the region inside C 0 . The condition for S *• 1 as r *• °° is required because a solution with S f+ 1 yields equivalent fields greatly different from the initial field.t. there exists i/>0 decreasing at infinity and satisfying the equation 3„[VM(Q. Generally speaking. We shall return to this subject below. and if there exist fields AM not equivalent to those over the region C 0 . these fields are located within the region C M .6 V. As shown in the appendix.e. integration can be cut off on the boundary of the region C 0 . there is a significant statement that for the functional integral. for such a field eq. then there is always a similar field equivalent to the former. even at the level of the things we can prove.
the difference between S and unity. Hence. equivalent to the field A^ = CM +#M. to be orthogonal with respect to ip0. ( ') = ~ ]ji SpJd4x{<p03M[flM)Vo] Vol^o]) > (26) TVVSpJ^d 4 * . Nonstrictness of the derivation due to the fact that we ignored the continuous character of the spectrum at e > 0 can be easily avoided.V.<PO] . i. In essence.s. (22) into (21) and taking into account (20). Gribov / Quantization of nonAbelian gauge theories Substituting the matrix S in the form exp a into (13) and confining ourselves to the terms quadratic in a. i. a«i^0 . and we shall not dwell on that. Using (24). <p0].S]=9M[aM. 7 (21) (22) Substituting (19).a]]=0.h. For this purpose..a]iaM[a[VM.e. we have found S.^ e a 1 SpJ d4x ^0dM [aM. we get 9M[VM(0. (27) which was to be proved.e. <I = 0 M + LV M (0. for a vanishing solution it is sufficient for the r. The solution to (21) may be sought for in the form a = >p0 + a. imposing the orthogonality condition of the eigenfunction <p0 on the eigenfunctions of other bound states together with eq. i. it is sufficient to calculate the shift of the level from zero due to the fields aM and a'^: e(a) = .e.N. we obtain 9M[VMC4). Sp/d 4 * ! ^ ^ . one can gain some insight into the cause of the field equivalence over . +V'O9 M [[ V M(0. (24).^0]]} = 0 . whether it is true that there exists a bound ghost state in one of the fields A^. (25) and clear up the question of whether A'^ and A^ lie on both sides of Cj. ^ ] i<MM^o[VM(0. and thus may now find the field A'^ = CM + a'^. A ^ and that there is no such state in the other. which defines the normalization <^0> w e obtain e(a)=e(a').v0]+^M^o[VM(0^o]] • (23) Since the righthand side of (23) vanishes at infinity. (24) This equation defines the normalization of i^0. The derivation can just as readily be repeated for the fields close to any £„.
(28) close to Hl.e. 3 the regions C0 and Cx when considering the effective Lagrangian in (15). J? has an extremum along [VM(C)> ^ol d does not change upon replacing ip0 by iPo. Clearly e ^ J ) = eG4~). As will be demonstrated in the appendix. but makes it more difficult to understand the physical content of the . such intersections actually occur. 3. Since the direction [VM(C% i/'ol is generally not orthogonal to the curve 2 j . the distribution of the equivalent fields can correspond to that shown in fig. The property of the fields CM lying on the line 5 j .V. is that there exists a solution to eq. takes the form £(A) = 2(C) . Hence.Vo]]} .4* = CM ± [V M (0>^o] a r e equivalent. 5. Gauge nonuniqueness and limitation on the integration range over the fields in physical spacetime So far we have discussed the functional integration over nonAbelian fields defined in fourdimensional Euclidean space. an (30) i. Gribov / Quantization ofnonAbclian gauge theories Fig. where the equivalent fields are located on dashed lines in the opposite directions to £j. Consider now the fields of the form ^M = C M + [ V M ( 0 > </>o] • (29) Then the Lagrangian £. S = ~ ip SP V " " " ^ SpO M ^ M ) 2 .N. this will result in the existence of at least pairs of equivalent fields within the region C 0 .\ S P { [ / > ( 0 . ^o] [>>(0> Y>o] + 2F M y (0[[V M ! ^o]tV y . these equivalent fields turn out to be mirrorreflected ones that are always equivalent. This somewhat simplifies the mathematics. and in this sense the equivalent fields do exist in C 0 .From this it follows that the fields . to secondorder in I/J0 . 3. but in the examples considered. the field doubling thus obtained is independent of the field magnitude and insignificant in the functional integral. (20). If structural lines can intersect as shown in fig.
and defines the ghost transition from the state with negative energy to that with positive energy.V. respectively. Now. (15) for the action. (31) where ^ ( D ) indicates that the integration is performed only over the region C. Gribov / Quantization ofnonAbelian gauge theories theory and leaves a feeling of dissatisfaction related to the need for analytical continuation of the results. This indicates that equivalent trajectories will only occur through those solutions of eq. disregarding the possibility for the equivalent fields to exist in C 0 : W = fe~f£d*x B(4)<V(a) &4 . In this case. The difference is in the real form of eq. The effect of the field magnitude restriction on the zeropoint oscillations and interaction in the lowmomenta region In this section we shall try to analyze how a limitation on the integration range over the field in the functional integral affects the physical properties of nonAbelian theories. Since the ghosts are quantized in the same way as fermions. .N. The restriction of the integration in the functional integral to the region C 0 implies the restriction to the fields in which no classical ghost formation occurs because the formation of ghosts merely redefines the fields A^. respectively. (13) which have only negative and positive frequencies as t > —°° and +°°. then A'^ — A^ as t > —°° and +°° should contain only negative and positive frequencies. according to Feynman. eq. we integrate over fields A^ which have only positive frequencies at t > —°° and negative frequencies at t > +°°. Certainly. these conditions play the same part as those at infinity in the Euclidean case.4M = 0 because of frequency conservation. such an equation has a solution under the boundary condition specified above. and the linear equations will have no solution at. (13) result in the fields 4^ which differ from A^ in pairs of the gauge quanta produced. (13) defining the number of equivalent fields. One of the ways for eliminating this nonuniqueness is a changeover to Euclidean space. the general statement that the integration should only be performed over nonequivalent fields is independent of the nature of the space. if we want to have two equivalent A^ andyl^ under the same boundary conditions. Clearly. and formula (15) holds. 9 6. Such solutions will exist for nonlinear equations or for sufficiently large fields A^. /. this equation is a hyperbolic one having nonzero solutions even in an Abelian case (S = exp <> y is an arbitrary solution to a wave equation). In a similar manner it can be said that solutions of eq. (18) at e = 0 is one for the ghost wave function in the external field A^. If the field A^ is situated on the line C t . For instance. We shall proceed from eq. the process is apparently interpreted as the classic formation of ghost pairs in the external field. In normal spacetime this nonuniqueness does not show itself because.
consider the Green function of the FaddeevPopov ghost G(*)=77 f e ^ £ t l 4 * a  .^(D) will be the cause. The fact that there are no other causes for the interaction cutoff is equivalent to the statement that without ^ ( D ) zero fluctuations of the fields tend to leave the region C 0 . we may say that we study the contribution to the functional integral close to the curves £„ when calculating G(k) near the "infrared pole". Since the influence of longitudinal fields cancels in the calculations of gaugeinvariant quantities. 2 48TT v .. where^(D) is insignificant and perturbation theory works. if we calculate G(k) in perturbation theory. omitting "^(D) and expanding over the coupling constant.N. . and hence°l? (D) is definitely significant at momenta below or of the order of the "infrared pole" position. with k2 below the singularity position. whereas at large k we are within C0 (low A). For this purpose. "^(D) makes it impossible for a singularity of G(k) to exist at finite k2 because. From the standpoint of (32). Higher corrections [8—10] and instantons [11. perform the integration over A in (32).e. G(k) can be large only due to the integration range for the fields where • is small. 48TT2 71X'l3J22)(3J2::^Jr\ » 2\ (3/22)(3/2a/2) k21 (33) where A is the ultraviolet cutoff. Both things would indicate that • has ceased to be a positively defined quantity. close to the lines 8„. Such a possibility would indicate that at k2 = 0 we feel the fields on the line fi. If no other causes are found.12] only bring it nearer.10 V.In —. we get G(k)= ~2 \ VTY^. i. i.e. It is interesting that transverse fields (low a) act on the ghosts as attractive fields and longitudinal fields as repulsive ones. will correspond to the real vacuum if ^ ( D ) is taken into account. let us see whether the restriction ^ ( D ) is significant from the standpoint of what we know from the perturbation theory analysis.e. i. connected with the singularity of k2G(k) at k2 = 0..e. Furthermore. Hence it appears quite natural that the fields closest to the boundary of the region C 0 . /A 7T. i. Gribov / Quantization ofnonAbelian gauge theories First of all. 2 2 2 2 k where (33) still holds. G(Qwould either reverse its sign or become complex. we have left the region C0 when integrating over^ M The only possibility that now remains is that k2G(k) has a singularity at k2 = 0. (32) It is well known that.. Up to now. a is the gauge parameter in £. ^  A : >   n i m Q ) c L 4 . From this it is obvious that G(k) becomes large at a < 3 and physical k2 (in the Euclidean space) such that \\g2C. all attempts at finding the mechanism for removal of the "infrared pole" have not been successful.
we choose a transverse gauge (a = 0). We write Gaa(k. that k is conserved in a typical field of zeropoint fluctuatioHS ((k\ 1/Dfc'>fc'=fc is proportional to the volume of the system which is replaced by 5(k .a(k. (34) The firstorder term gives no contribution to the diagonal element.A) = + ! + 1 1_ + — . (35) A^(q) is the Fourier component of the potential A^. On averaging over the gluon polarization directions X.x(q)\2 over the main range of integration with respect to q decreases monotonically with q2. A) < 1. we have 0(!l A) ' 4 J(2T04 (kq)2 V k2q2J (3?) If \Aa. In this case it turns out that there appears a characteristic scale 2 2 K 2 defined by the condition g In A 2 / K 2 ~ 1..k') after averaging). of course. If it were not for the roughness of the calculations and difficulties with complex singularities of D(k). but unfortunately we have not yet succeeded in doing this. The ghost Green function under k2 * 0 isG(fc)~C/A:4. as will prove to be the case in what follows. The secondorder term is V F fd^ Al(q)Aav(q)k^kqX_ PUr^Y* ^T^ 2 h*J{2iif (kq) A V =T2 0ik. A). For simplicity.A). so that at k » K2 the gluon and ghost Green functions remain free. All we could have done was to write this criterion to second order in perturbation theory and then calculate the functional integral taking no account of the interaction except for ^ ( D ) . a(k.N. a\ l/D\k. Gribov / Quantization of nonAbelian gauge theories 11 For checking the above by a specific calculation. A) decreases as k2 increases and hence as a nolevel condition use can be made of <M)3/&^L'<. „ • k2 1 . a) in the form of an expansion in perturbation theory (where a is the isotopic index) < i i Gaa(k. A) defines positions of the poles G(k. then o(k. The nopole condition at a given k is a(k. The gluon Green function D(k) has complex singularities and is nonsingular at k2 >• 0. A) = —{k. V the volume of the system. if any. to a second Born approximation since GQcA)*^ J . this would be the right thing for the colour confinement theory. „„. A) V_ (36) In this case we assume.V. Let us show the way this is obtained. . one must write ^ ( D ) in a constructive way.
(46) _f. . (39) J 2m0 J X exp •^B^i^tf^ gl V 2 '"* "* W>/i T .a \ (k2 + —)\AKa(k)\2 k ] . with F*o°weget § r f£JL _ 4 ^_ = «if22 J/ (2TT)4 < 7L K 4 = 11 ^ T + „ (43) 3« *2In^=l. (42) Setting (30g2/V= K4 . the gluon and ghost Green functions are D$(k) = Al(k)Abv(k) .1 + — = 0 . A2 (44) If the saddlepoint value j30 is known.X. GribovI Quantization of nonAbelian gauge theories Taking (38) as a condition for 0^(0).4.^ ) ^~A . " \Aa'x\2 1 ql (40) where Kis the volume of the system. W is W= JdAexp . replacing i? by i? 0 in (31) and omitting II D. with the saddlepoint value j30 determined by 3" g2 V^ — — Li —„ 1 1 =— . we get « being the number of isotopic states. if 0^(1 — a(k. A)) is written in the form ^ ( 1 . In this case. substituting )3 = /30 in it and omitting the integration over (3. The integral over j3 can be obtained by the steepestdescent method.\ g £ k.A))= f^r Sl°«>'A» . instead of (31) we obtain a functional integral which is easy to calculate.o(0. we can return to the functional integral (40). 1 ° ~k2(lo(k)){ r a {) =8 a b ^ .N. _4g2 rd4q k2H2*)4 k^DZti2) (kq)2 . . (45) Consequently.12 V. so as to obtain an effective functional integral for calculating the correlation functions of the fields . Calculating the integral over A.
The fact that the significant range of integration in the functional integral turns out to coincide with the boundary of the region £]. 6 we discussed the effect of limiting the integration over the fields on the properties of vacuum fluctuations in the invariant Euclidean formulation of the theory. in this section we shall rewrite the foregoing analysis for the Coulomb gauge [13] where the Green function of the ghost determines directly the Coulomb interaction.a)"7!? (1 — a). the functional integral W incorporates the fields which satisfy the threedimensional transversality condition ^ = 0. In so doing. (50) .V. but the ghost Green function in an arbitrary gauge is not connected directly with the Coulomb interaction at large distances.. the last term in (42) has no effect at V* °° and hence ^ ( 1 — a) is equivalent to 5(1 — a). we adduced arguments for singularity of the ghost Green function as k2 + 0 (for example. The arguments for singularity of the ghost Green function hold here as well. (49) and momenta TT. This certainly is an indication of a substantial longrange effect in the theory that may result in colour confinement. In this case. As k2 > 0 327TK2 (48) in accordance with the above.which are canonically conjugated with them and stand for the transverse part of the electric field Titf^tVMo])1. Hence. The most natural way of formulating the Coulomb gauge is the Hamiltonian form which shows up vividly the unitarity of the theory because of the lack of ghosts. To this end.r) Q + K (kq)2 2 13 / \ (fe?)2' (47) spectively. l/£ 4 ).N. however. 7. is evident without calculating G(k) because. which is equivalent to an attempt at taking into account the effect of the determinant in (31). Gribov / Quantization of nonAbelian gauge theories 1 (fd4q k2^2kq 1 2 4 4 4 g nk \J (2. when calculating the saddlepoint value 0O. We would obtain the same result when calculating with the function (1 . due to (43). a singularity of the ghost Green function as k2 > 0 of the type 1/fc4 is indicative of a linear increase in the Coulomb interaction with distance. We shall show that the situation which involves a restriction on the integration range over fields and a cutoff of the infrared singularity found in perturbation theory is exactly the same as in invariant gauges. Coulomb gauge In sect.
(57). (51) <X(it„ At) = .dj<p].N. (55) a zero eigenvalue of the operator A(A) defining the Coulomb potential according to eq. as for invariant gauges in (51). (13) [V.S + ]=0. The number of fields equivalent to a given field Aj is determined by the number of solutions of the equation similar to eq.Aj 9i (77. Let us discuss now how a limitation by the fields within C0 affects the spectrum and zero oscillations defined by eq.we set 4 f w Q K 3 TJ \Ax'«(k)\2 hi < j (58\ . (52) (53) In this case. (53). Ad] d 4 d 7 7 . (57) bearing in mind that 9i is determined only for the region C 0 . instead of the functional integral we can use the Schrodinger equation 2 fd3x 9C (ig ^ .*} . {TT? + H2 (Ad + 3. In the Coulomb gauge. As a result. A(A)<p=p. A)J MAd = E^(Aj). the functional integral takes the form W= fexp \ J V * {TT.*3. (54) The condition for the existence of two equivalent fields is the existence of a solution of the equation [V/(4).. we draw a conclusion that the integration range in (51) should be restricted to the region C0 where the operator A(A) has no eigenvalues. This region coincides with the one where the Coulomb energy density VCoul=l2pld2~P (56) A A does not go to infinity anywhere except for its boundary.3flp]=0. As before. the integration should be performed over gaugeinequivalent transverse fields. GribovI Quantization of nonAbelian gauge theories The integral overyl 0 is calculated for fixed Aj and cancels the FaddeevPopov determinant.C4).. A(A)^=[Vj(A). ti1=\F.4.S9. Repeating the arguments given above for the fourdimensional case (see also the appendix). for the nolevel condition in the field .jE„..7ti].236 14 V. p=[Ai.
where K4 is the variational parameter. we get an oscillator equation for/with k2 + n4/k2 instead of A2. we obtain.q k 2 ^ <x .\.q = k . Therefore.k..q • (61) Calculating the energy minimum for the system with the fixed value of the lefthand side in (60). A2 (63) The ghost Green function is G(k)=(k\i:\k) A 1 rd^ (k2 2kk')(\ (kk')2/k2k'2)\ k24g2n J(27T)3 k'(k'~k)Wk'*+K* j as A:2 > 0 G(k)= ' [b ' 5g2nkUn(K2/k2) • (65) . for free oscillations \a\iQ(k)\ ~ 1/A: and hence the lefthand side of (60) is infinite. instead of (57).Ajd2Aj).q(k)\2 3V k. (60) Taking no account of (60). Omitting in 9C(it.(*)+*M. Gribov / Quantization of nonAbclian gauge theories 15 obtained in a similar manner from the Green function. Eqs. In this case.q(k)) k. This means that free oscillations correspond to the fields far outside the region C 0 .(*)}lK«)=W). an equation for an oscillator system £ k. Kq {<.\.N. (59). (60) can be solved approximately by the variational method assuming that * = 11 f(ax. but as this takes place. A" + ft ys T „4 (62) The energy is at a minimum with « = 0.A) all terms except for 9f0 = —\{n2 . the groundstate energy and average squares of oscillation amplitudes will be E\ \aKq(k)\2 S k. K is determined from the condition 8 (2TT) rwrzi *g J0^3 A : V F T ^ ~3TT l nK< =l • * ^ zr 3 2 2 r d3k _l 2g2 . the lefthand side in (60) equals infinity. (59) provided that V £ \ax.237 V.
+/ 3 (r)An.2) AiQc) goes into At(x) = S AfS + S dtS so that / . Finally. which is indicative of a decrease in the zero oscillation amplitudes for momenta below a particular value and of a linear increase in the Coulomb potential with distance. Polyakov.16 V. = / . Let us begin with the case of the threedimensional space (Coulomb gauge) and the group SU(2). Gribov / Quantization of nonAbelian gauge theories Thus... are Pauli matrices.1 . cosa + (/ 2 + ^ ) s i n a . (r) ^ + f2(r)n ~. A. (A. L. this crude calculation shows that again there is a characteristic scale beyond which zeropoint oscillations become small and the Coulomb potential increases linearly with distance. n2 = .l) n = inaoa. considering that the number of zero oscillation modes pushing the system outside the region C0 is infinitely large and that the boundary equation comprises one condition (60). giving the simplest examples.e. f3=f3+\a.N. we have no right to ignore the Coulomb energy which may go to infinity on this boundary. Mildxt = h [fi + (2/r)/ 3 . we have good cause to think that the system will remain close to the boundary despite the Coulomb interaction.2fjr The condition dAildXi = dAilbxi (A. We consider "spherically symmetric" fields Aj.L. Under the spherically symmetric gauge transformation of the form S= exp{\a(f)n} = cos(ja) + ft sin(^a) + + (A.4) . In conclusion. Appendix In this appendix we consider the properties of gaugeequivalent fields with equal divergence. Dokshitser for numerous most helpful discussions. In this case. (A. However. it should be noted that since zeropoint oscillations of the fields in vacuum turn out to be on the boundary of the region C 0 .N.A. the condition (60) will be satisfied all the same. Lipatov and Yu. aa. the fields dependent on one unit vector nt = xt/r (r = \Jx}). fi + 5 = f\ sin a + (/ 2 + ±) cos a .M. I wish to express my sincere gratitude to A.5) 2 (A3) ]. The general expression for such a field has the form At(x) = / . i. Belavin.
(A. (A. If the forces/. If sufficiently large. then for a solution to exist at finite T. However. horizontal force 4 / .8) decreases slowly at infinity. Then. otherwise the field A x is singular as r » 0) and tend to zero as r > °° (T >• +°°). (A.5).6) reduces to the equation for a pendulum with damping in the field of the vertical force 4 / + 2. The field with a corresponding f2 lies on the curve £j independently of . as r >• °°. In order for eq. as r > —°°. and the force perpendicular to the pendulum. . and f2 are equal to zero as r »• 0 (r * »>. (A. Let the forces and initial conditions be such that throughout the whole "time" ~°° < T < +°°. —4f\ a + d .6) If we introduce the variable T = In r. (A. upon executing a number of oscillations in the field.KM GribovI Quantization of nonAbelian gauge theories 17 .9) to hold. the pendulum should be in the position of unstable equilibrium. In such an event. if its initial velocity as T *• —°° is not specifically selected. we obtain the equivalent field At which decreases fairly rapidly at infinity.9) should be satisfied. and the equivalent field Ai=nMldXi~ \jr . (20) for the threedimensional case. eq. In this case.cos a) = 0 . Such a solution corresponds to S > n.7) From this analogy the general properties of solutions to the equations of the equivalence conditions (A.( 4 / r 2 ) { ( / 2 + i ) s i n a + / 1 ( c o s a .oL / •/ I 1 Fig. (A. a(r) « 1. to a zero approximation the equation a + d . This equation is simply eq. the pendulum starts damping and once only the vertical force remains. We consider several versions of such a possibility.( 2 + 4 / 2 ) s i n a + 4 / 1 ( l .6) are readily seen. —// 2 dr should have a particular and sufficiently large value. a = 0. it comes to stable equilibrium. these forces can under specifically selected initial conditions restore the pendulum to its unstable equilibrium position.l ) } = 0 .2 a ( l + 2/2) = 0 (A. 4 is equivalent to the equation a" + ( 2 / r ) a ' . exceptional cases are possible.
f 2 ) . there are no secondorder terms and a solution exists only for/ 2 larger than the value required for a zero energy level.7) in the external force 2A/e 2T on the righthand side perpendicular to the pendulum. It is obvious that in this case there are two solutions that differ from one another by the sign of a.. deep inside the region C 0 . it is convenient to deal with the group 0(4) from which SU(2) is trivially separated. ( / .e. This possibility may exist even with/ 2 = 0. For example. the solution comes into play through choosing the initial conditions. Such solutions exist at large enough/. ( . . reflected field obtained f r o m / . The amplitude of a is determined by taking into account thirdorder terms. should/. / 2 . .as antihermitian matrices for infinitesimal transformations in the group 0(4). The equation for a corresponding a(r) will differ from (A. we have an inhomogeneous linear equation for which the choice is made in a trivial way. let us discuss the question as to whether for a particular field At an equivalent field A\ with a specified difference «A/in their divergence can always be found. i./fy)..For constructing a scalar.7) enables us to get a solution with a somewhat larger or smaller f2 (within or outside the region C 0 ) and to determine the amplitude of a as demonstrated in the text. i. eq.e.3) because a in a complete revolution changes from 0 to 2n./ 3 . In this case. a2 (/.. we . . with/ 2 = 0. as we have seen. / 2 . leads * to an equivalent. Instead of />. f3.18 V. . It is easily shown that the fields / . Taking into account the quadratic term in (A.This situation corresponds to the phenomenon of intersection of the equivalent field lines discussed in the text and illustrated in fig./ i . = 0. and/ 2 are large. one may choose o^v = \ (7^7. / 2 . If/. / 3 ) are outside C 0 anyway. (A . / 3 using the solution a .. and/ 2 close to Another possibility for the pendulum to regain its unstable equilibrium position is to execute one complete revolution (or more). for small/. 3.. both reflected fields are obtained from one initial field. ./ 2 and A/ be small.9) has an infinite number of solutions) and the fields situated in C„ with a finite n. = 0 lies on the intersection of such lines. The second solution in the field . — 7. Finally. the fields inside C0 have their equivalents of two types.. The field for which/. Hence the functional integral should be divided by two. . = a ( / .. a solution with a tending to 2nir as r*°° because. . f 2 ) .N. Hence./ 3 ./ . Two other solutions a. independently of the magnitudes of the fields/. f2 and —/. With/. / 2 ) and a 2 ( / i . Accordingly./ 2 is of the order of unity according to (A.8) (a »• n) and lying within the region C^ (it is easy to see that with/ 2 = 1 as r* 00 . f2) and < 2 = . Gribov / Quantization ofnonAbelian gauge theories the values for/! and/ 3 .6) always has two solutions a. but the equivalent fields corresponding to them differ greatly from the initial field and are likely to lie outside C 0 . the fields which possess asymptotics at °° of the type (A. it is likely that there exists. = 0. We now turn to the fourdimensional space. It can easily be shown that they are in the region C0. —f2 are gaugeequivalent and have equal divergence./ ..a ( . (A. The occurrence of two equivalent fields in C0 in this example does not point to the necessity of introducing A ^ ) in the region C0 because it shows the symmetry of the problem with respect to reflection. which do not matter. / 2 ) . In this case. "almost" without exception. if/. / 3 . .
Belavin and A. Popov.e. Jones. Politzer. Yu. 1 0 ) VI (nala)2 where HM =X^I\/X^ and /M is a constant unit vector. its structure is much the same as that of (A. r (1 rii) 2 n 4 (A. p. and we do not see any reasons why the structure of its solutions should differ markedly from (A. Lett. 12) The transformation formulae between//and/. 773.«. Khalatnikov. Gross and P.6). Phys. Phys. 110} W.D. Rev.N.S. Callan. The equivalence condition is 3 2 ff. [Ill A. Abrikosov and I./ 1 ( l . preprint C002220115 (August 1977). Landau. 30 (1973) 1346.coincide with ( A 3 ) .N. Wilczek. 1. Gribov / Quantization of nonAbelian gauge theories need an antisymmetric tensor. 160(1967) 113. (A. 112} C. Migdal. Vol. Phys.. Lett. (A.D. Feynman. if a is replaced by/J.^ M ( A 19 . 33 (1974) 244. 25B (1967) 30. r=\fxl. B75 (1974) 530.M.J.A. 1117. Phys. Pol. Lett.11) The field A^ which preserves its shape under this transformation has the form A^Ad^+Md^ + ^d^ .6) which is dependent on one variable p 2 = r2(\ — nf) and has the same asymptotics.B. Nucl. Faddeev and V. 1977.13) With/i = f2 = 0. 24 (1963) 262. Phys. i. n.M. Lett. ZhETF Pisma 19 (1974) 317. (6) I. Despite two variables. Rev. 30 (1973) 1343. Rev.6).  5  D. 4] H. Materials for the 12th LNPI Winter School.A. = / M « M . References (11 R.2 [ ( / 2 + l ) s i n / 3 . 113] V.J. 59B (1975) 85. Dashen and D. Rev.N. Caswell. J7] L. 131 L. 147.G. Tyupkin and A. DeWitt.R. The gauge transformation matrix between such axially symmetric fields can be written as S= exp{lj3(r. This indicates that the field cannot be spherically symmetric. Princeton Univ. 162 (1967) 1195. Phys. A. Polyakov. ZhETF (USSR) 10 (1969) 409. Lett. Acta Phys.T. at least two vectors are required. Gross. Belavin. Schwartz. Phys.)!//} = cos \p+ t// sin \$ .1293.A. {2] B. Gribov. 9J D. . Khriplovich. R. A.A.P. DAN 95 (1954)497.D. (8 A. It can be axially symmetric if we choose as antisymmetric tensor ^ . ^ = 5 0 M „F M „.c o s  3 ) ] = 0 . there is a solution similar to (A.V. which make this equation more cumbersome.
A. BECCHI University of Genova Italy A. VeloandA.GAUGE FIELD MODELS C.S.Z.) renormalization scheme has been stated in Stora's lectures \j>~). .P.A. Renormalization Theory.R.P.H. expressing in terms of the Green functions the quantum equivalent of the Noethertheorem are affected with quantum correction depending on the particular renormalization rules. 269297.P. . Copyright © 1976 by D. proved in the framework of the BogoliubovParasiukHeppZimmermann [ff] (B.A. This weak form of the Q. DordrechtHolland.g.P. in those which are based on the dimensional regularization of the Feynman integrals) these quantum corrections seem to satisfy a not yet well defined criterion of minimality. S.P. version of the Q.H. Reidel Publishing Company.Marseille Lectures given by C. ROUET MaxPlanck Institut fur Physik und Astrophysik MUnchen R. (e.) have been among the most important subjects of the first week of this School [2. which can be deduced on a very formal ground. In some "clever" renormalization scheme [3].Z.j. STORA Centre de Physique ThSorique C.N. Wightman (eds. All Rights Reserved. Stora has been able to prove rather general results on the renormalization of lagrangian theories with broken symmetries [/}. The purpose of these lectures is to continue Stora's G. Exploiting the B.3] . It turns out that the naive identities. BECCHI INTRODUCTION The renormalized versions of the Schwinger Quantum Action Principle \J\ (Q.4]. In a more general framework they have known power counting properties [2.
 analysis discussing with the same technique a more sophisticated class of lagrangian models. We have a model depending on a finite number of parameters and we have to show that the original physical properties of the theory are not affected by the introduction of the new "unphysical" degrees of freedom. Hence it is of particular interest in the construction of renormalizable models based on vector intermediate bosons for the weak and electromagnetic interactions L8] . Hence our first JJ goal will be to prove that one can fulfill the Slavnov identities to all orders of the renormalized perturbation theory.). In much the same way as in Q. in which the gauge group is semisimple the gauge photons must be massless except if the gauge symmetry is spontaneously broken according to the HiggsKibble (H.) mechanism I 9J. mechanism.K. B IXC HI r T AL. photons and the FaddeevPopov I 10 • ( ^ .K. (Here by renormalizable we mean satisfying the necessary power counting prescriptions.).M. In order to avoid infrared difficulties we shall discard models in which massless fields are involved. We shall restrict ourselves to models involving . (Q. under a set of nonlinear field transformations (Slavnov transformations) I 12] explicitely involving the cj). the Slavnov identities 1L . limiting ourselves to the analysis of theories in which the gauge symmetry is completely broken according to the H. In the G. This procedure completely changes the renormalization problem.D.M.) The classical way out of this difficulty implies the introduction of a certain number of degrees of freedom which are not physically interpretable (e_.F. which.T T . g. that a gauge invariant lagrangian model cannot be directly quantized apparently preserving the local structure of the theory and leading to a renormalizable perturbation theory. this is the consequence of a set of WardTakahashi identities. express the invariance of ' the FaddeevPopov lagrangian of the G.M. that is the nonabelian (YangMills) gauge field models (G.A. " " \ ghost fields. It turns out clearly from the study of quantum electrodynamics. Then we shall discuss in some details the gauge independence of the physical scattering operator and we shall outline the connection between its unitarity and the Slavnov symmetry.13].F.270 C. the oldest and best known gauge field model.D. This is the only known class of renormalizable models involving a self interacting system of vector mesons. scalar. the longitudinal.P..E.) ghosts). More precisely we have to prove that the restriction of the scattering operator of the theory to a certain "physical" subspace of the Fock space is unitary in the perturbative sense and independent from the properties of the unphysical particles. The complete combinatorial proof of unitarity will not be examined here since it is discussed in details in the referred literature I 12.F.E. owing to the Q.
An infinitesimal gauge transformation of parameter SOJ^^OC) is defined by / • Sit (*).LS l. involving semisimple gauge groups.P. 2. 7."' ' ' ^ ' * <Xa4. model.13.A.i3 be a multiplet of scalar fields and Quvt 0134. In section three we study the renormalization of the Slavnov identity in the G. Let and Wo.2. Section four deals with unitarity and gauge independence of the physical S operator in the SU(2) H.} ' i1T*'<r ) With Q 3 .](x)=.^ jj*? &V»> .M. This will greatly simplify the analysis of the possible quantum corrections to the Q. THE CLASSICAL SU(2) MODEL In this section we shall examine in some details the classical structure of the SU(2) HiggsKibbleEnglertBrout model and the problems which are encountered in its quantization. In section two we discuss at the classical level the algebraic properties of the SU(2) HiggsKibbleEnglertBroutFaddeevPopov lagrangian and we exhibit its invariance under Slavnov transformations.2.K.F. which in technical terms will be reduced to the study of the cohomolqgy group of the Lie algebra characterizing the gauge theory ..  [ e« M fay*. +(<r.D 271 semisimple gauge groups.GAUG1 l ' l L MODI. For what concerns the basic tools and definitions (in particular the functional method) that we shall extensively use in the following we refer to the first section of Stora's lectures.3 i a multiplet of YangMills photons.F)to.
(5) it turns out that the mass matrix of the scalar fields : (6) has three null eigenvalues corresponding to the Tl^ modes and a positive eigenvalue corresponding to the O" field. To discuss the quantization of this model we have to examine the wave operator corresponding to the free part (that bilinear in the fields) of the lagrangian. From the invariance property of H and from Eq. This result expresses the content of the Goldstone theorem £14].272 C. BK. A lagrangian invariant under such transformations is constructed in terms of the antisymmetric covariant field tensor : and of the covariant derivatives : it has the form : 2. The parameter F plays the role of the O" field vacuum expectation value. In particular the free lagrangian of the photon — TT channel is : . 2 Here H is a polynomial in " H^ f\0+1) 3 [ V + > J invariant under the transformation (1) and satisfying the condition : \ M)U=° H (5) expressing that the classical vacuum corresponds to the configuration tC s O .CCHI tT AL.
It involves.v f e (8) exhibits the Higgs phenomenon 9 (the photons are massive).<)b [ f*.GAUGE FIELD M O D E L S 273 The Fourier transformed wave operator factorizes in two submatrices.D. to which "physical" particles are associated. The first associated with the transverse modes of the photon: ^V) = ^(9 ^V. Another solution is based on the FaddeevPopov lagrangian I10 I. The lagrangian is : < (%. In the cases in which the theory is spontaneously broken one can eliminate the degrees of freedom with degenerate wave operator by means of a field dependent gauge transformations 15 1 . beyond the scalar and vector fields.M] with c ( c (10) .T) \. lagrangian.<. the transverse modes of the photons and the O" field. It follows that our invariant theory is not directly quantizable. is : 1 =£ M 4 \ ^  ^ 1 2 m f (9) The determinant of this matrix is identically zero.F. two conjugate multiplets of scalar fields  C * \ / S *u J quantized according to the Fermi statistics in a Fock space carrying indefinite metric . «^)= *~<x. However the resulting lagrangian is not renormalizable by power counting. This is the natural extension to the case of non abelian G. of the usual Q.TT channel. corresponding to the coupled longitudinal photon . This reduces the degrees of freedom of the model to a minimal number.M.E. The second.
E. This generalizes the well known Q. This new lagrangian can be straightforwardly quantized in a Fock space with indefinite metric .TT. 8. which at the classical level expresses the invariance of the lagrangian under the following set of infinitesimal transformations (Slavnov transformations) J odifAH) r 3 topi*) Sa3jil^) p _ d4) .= ^ ^ + ? ^ and (11) (<mc) 4 (x)Jdjgfc^ c f (j) B J StytV (12) . the corresponding perturbation theory is renormalizable by power counting. lagrangian we have increased the number of the unphysical degrees of freedom of our model. TV.74 C.term of the lagrangian.D. supplementary condition. In the tree approximation this is true if the matrix elements of the gauge operator ^ j C ^ ) within physical states vanish. Indeed now : However choosing the 4>. We are then left with the problem of proving that the restriction of the S operator to the above defined physical subspace of the free field Fock space is unitary in the perturbative sense and independent from the parameters specifying the M5. BICCH1 IT AL. We are now going to show that this follows from the Slavnov identity.
.l)« ' ' ^0) . . Thiis fact can be translated by means of the Q. C ^ \ respectively yields: 0= 32c (2. These identities are easily expressed in terms of the connected Green functional Z r6j .GAUGI: I ' H D MODl'LS IL 275 O A is a spacetime independent infinitesimal parameter commuting with S C^a ^ and £<luai\ but anticommuting with the "flTghost fields. "P. and for two transformations labelled by o•A^ and o A^ t O •XA >d ar o A . hence. 30. anti commute. . d*\ .t / t i* \ \ for the fields ^ ] = < TT^ (<T.< <fa. The Slavnov transformation laws can be cast together introducing the functional differential operator : 4= 4 * /Sir/ "fo ^o. iose the — < X Since. C«< . C^ 7 S4J = 5A<S^ d6) The invariance of oL under the transformations (14) can be checked immediately taking into account the composition law of the gauge transformations : [ s SG^X)' s Scoots) (17) Let us now introduce the system of external fields (18) whose couplings are defined by th new lagrangian . 0u. ^ % ' " 3 9d. owing to the group structure underlying the Slavnov transformations. Upon introducing the sources £ J j ..P. C . \ are invariant. also oL is invariant. (X) (15) Then the trans formation law for the f i e l d is : ) ty T .$ Jot. r 5c." "P ~P.A. at least in the tree approximation. the polynomials 1 Pri . into a system of identities relating the Green functions of the theory whenever the possible quantum corrections are absent.
CX'"* .(15) and 2Lfc»Kvs commute. reduction formulae it turns out that we can write the restriction of the S operator to the physical space in the form I 61: e^f^cO. Now the previously mentioned property of the physical fields implies that the operator .i)fcC3.(14) we see that the Slavnov transformed of the physical fields do not contain any term linear in c .(14) we see (23) ^J*)=(WC) M ^ C x ) Hence this property of the physical fields is not altered if we combine them with Q From the L.S.^L^sZ where C . BF. Comparing with Eq.CCHI ET AL._ Z(3. since from Eq.^) ' (24) are the canonically quantized asymptotic fields and the corresponding wave operators.Z.276 C.(10) and Eq. K.i)=zt(J. K. The corresponding identity for the vertex functional r(¥. It follows that the field Q is decoupled from the physical states : ^ . an amplitude containing gauge function that : Q* ° 3 Yk>Kvs as a vertex * s o n e par ticle irreducible with respect to cuts between this vertex and the . In terms of Feynman diagrams this property can be translated into the following : if "U'pW'JS ^ s a n y physical field.l*+F))<*> J i ^ stf sff1s& s £ a r * W *J> <21) (22) Let us now recall that we have defined physical degrees of freedom of our model the transverse components of the photons and the <T field.S < defined in Eq. since 2~ t>Vius amputates on the mass shell the physical legs. .
. U \ .. e~jr<. (26) 2.GAUGE FIELD MODELS 277 =0 This is the desired supplementary condition H=0 (25) This fulfillement of the Slavnov identity can be considered as the fundamental property of the G. x. c .TTcharge Q'*' : e^x^s.F.o ' (27) 2*'c . • *.c. For instance in Q. However to base our renormalization program on the Slavnov identity we have to verify that our lagrangian is uniquely determined (up to the addition of a divergence and a multiplicative renormalization of the fields) by the condition of Slavnov invariance.Q*ral. Q*<r. the most general dimension four lagrangian carrying null charge is : have d lmension non exceeding two.M.D. <2"*. J fields and If we assign dimension two to the external we introduce the <$>. \4>tr Q+"C. <?X . In the non semisimple case there may arise new terms. understood. u°. (J u . This is easily verified if the gauge group is semisimple.. The Slavnov invariance of the action is expressed in the functional differential form : where the polynomials ) <j . in the Stueckelberg gauge the mass term: G^+CC(X) is Slavnov invariant. the dimensional constraints ensuring the renormalizability of the theory bein<.K • Q"*.E.. .
BKCCHl UT AL. the general form of.*Mr£ _  XI< (35) . for example.(30)) writes : which is nothing but the Jacobi identity : r^r J " t r.  ay  and Eq. That is if a set of structure constants differs from another m e by perturbation. fields explicit. TT.o *> p &p T>SXTW (33) whose solutions are stable under perturbations I13. ot(u>) . .»^ + r.(14)) can be reconstructed. Now. concerning the external field independent part of the Lagrangian. Indeed. if i(<4>) is Slavnov invariant a fortiori : Making the dependence on the assumes the form : <$>. Looking in particular at the coefficients of the external fields we ge t : 4 £ * 4 < 0 « Z&^XS* f ^ is : =° (30) from which the structure of the Slavnov transformations (Eq. to the first order in the perturbation the two sets are equivalent.278 C. r 1 H .
^.(34) it follows that : L and t h a t : = 0 (36) j a ^ T n cJ60 (KC)OO=O O?) The general solution of Eq. if the gauge group is semisimple : . " ' symmetrical matrix.C^] (40) with \^tl = S ^ c £ } • Here \C* .. C„ } are the 4>TT. (mc) where yields : '*„/' : sa < 3 8.(37) is. lAtA 8 8 photons.F. In order to study the general case we shall use a slightly different notation labelling the quantized fields by : \^} = {4>i. By Eq. 3.M. satisfying the appropriate Slavnov identities. = V^. tne au e fields. RENORMALIZATION OF THE SLAVNOV IDENTITY In this section we discuss the existence of quantum extensions of G. with semisimple gauge group.GAUGI: l ' l L MORILS lD 279 where of is invariant under gauge transformations and the corresponding action is Slavnov invariant.(12) modulo a redefinition of gd and c. 0<c). Going back to Eq.(29) which is identical with Eq. We also introduce the sources: ISH^M and the external fields : (4i> . and J <0ay a system of matter fields.
: are the structure constants and a representation of the Lie algebra associated with the gauge group. charge and dimension assignements are the natural extensions of those in the previous section. linear in the fields: 9(x)»gd>.5 P "^ NT £ *rt and t . . r The coefficients Aand »*~. A¥ * .280 C.f & and Co) ^ C y (46) **(«•).tu. e O ? V ? ) 2 Mr rfrv 4 —1^1 6 b (47) . The purpose of this section is to show that the Slavnov identity : %+.M. ^tKi^Vi^l) *s a f° r m a l power series (45) Hence in particular : T \ (<V) .F.(x) »ith tf^s%l\ (43) A quantized extension of a G. hM^M The gauge function g* is chosen (2 4) The ^TT. BECCHI FT Al. is defined by means of an effective lagrangian : ^K^M^w^r?^^^ The lagrangian in t : f CO (44) to which the Zimmermann I 5 1 subtraction Index four is assigned.
can be solved in terms of know that : ^fff f° r an W y ^^ • • * • Now we .^X) is a polynomial in the fields and their deriva tives whose coefficients are formal power series in 'f : T _ «( ' V * h / V (x) * with V>A «4ef(.GAUGK FU.(48).(49) becomes : which is solved by : ! <i Kk )^° r T Now for a generic choice of the effective lagrangian the Q.P. h We shall proceed in two steps. (52) The ver where the righthand side means the insertion of the vertex LJ. ( i) into the appropriate Green function.(51) can be solved in terms of valent to show that the equation : 4 =°.LD MODLLS 281 can be solved in terms of the coefficients of otetC CM*'*7J.A. After a Legendre transformation Eq.. These coefficients are themselves formal power series in " . (53) i s equi To prove that Eq. the necessary condition : First we shall discuss Then we shall go back to Eq.{5Cj) tex = (Ajoc).(x) Vtm) fsj ( A.r) 3 (q. yields : B.
by a system of consistency conditions which to the first non vanishing order in •t assume the naive (classical) form (as in Eq.(*. Now. (3 _C QteCC ) sums up the quantum correction to the naive version of the Q.(52) the term of order t \ yields XV) (59) we have to the same order the consistency condition : (V) (V) $. J^tCt} * consequently /\ (jX.(X). The possible quantum anomalies to the Q. We shall f apply this method to the Slavnov identity and show that if the . we have •(¥.(x>rO (w) A / ^ (60) from which the absence from Z_> (pC) of terms of the form (57) follows immediately.(56) is fulfilled if Q fx.(60)). From a simple analysis it turns out that Eq.%) (5 5) where yJ.£*)= IU*)Q 6%) for some local functional ant Of'feff) °f dimension four. only for X smaller than V • Clearly Eq.255 282 C.A.P. (x). are constrained.$ (58) if Since picking out of Eq.to ApU)+D/>(3)A.I i W ^ Sff^'W +£.) do not contain terms of the form : where the symbol o> (2 means symmetry under the exchange of the two indices. and consequently depends on "SeCC.A. taking into account the anticommuting character of the operator Q.(54) can be solved if: £>. B1CCHI l£T AL. in addition to power counting. This is a particularly simple example of our general algebraic method. (v) f jy) (v) A.P.
.LD MODI LS 283 gauge group is semisimple. We now go back to the Slavnov identity.ti+ \ * V « M ) J W + (66) and d CP) ^ . due to Wess and Zumino fl7j. the consistency conditions are sufficient to ensure that by a suitable choice of the effective lagrangian all the possible anomalies can be eliminated in the absence of AdlerBardeen anomaly [l5].GAUGK 1 ll. Assuming from now on that the ^>.\* where ^C*.VJ (67) The remainder R C) v depends on the coefficients of Jiett{ty. (61) where the vertex : (62) has dimension lower than six and : (63) where : (64) Furthermore there exists an integer V2A such that : (65) ' and from Eq. Making explicit the dependence on the external fields we have : .TT.P. of defining the AdlerBardeen anomaly by means of a system of consistency conditions.A.^ > } } ^= <SA*) iS J.(49) we get from the Q. This is the full exploitation of the idea. "ij up to order V — 4 . part of the effective lagrangian is determined in such a way as to satisfy Eq.(63): (V) / v / ^ ) (o) iV) \ A c*)= .
TT. First we introduce a new external anticommuting field J^(x) carrying 4>.284 C.(49) is satisfied. A ( X ) is constrained by a system of consistency conditions which we are now going to derive. BliCCHI I T AL.(A2^ (.lifi+(£*l (*•*•*> (71) In the righthand side of Eq. To order V these new terms can only arise from the naive variation of A ( X ) since the 6 dependent radiative corrections appear to an order in "n strictly higher than V (that of the new coupling). ^x). charge Q*W= 4and coupled to the vertex [Slj [_A(x)J . In the theory defined by the new lagrangian : the Slavnov identity writes •J MUM . Hence we can write A*1.A>+[^r]w+[^>) with : (68) A» .(71) the insertion /\ lumps together the quantum corrections involving S couplings. +\ (*/«£ff) (69) This is the structure of the possible anomalies to the Slavnov identity. However if Eq. fax j ^ ) [ L ^ e $ « W ' T r + ^ + *M+ (72) (73) .
and — A .i4[4ifl^*o since : i S2 =0 dxa 3 <SjMx) Jj&^J (75) vanishes owing to the anticommuting character of the external field To order V Eq.(74) yields the desired system of consistency conditions for & : (76) The first equation ensures that the system .GAUGE FIELD MODELS 285 Now we can compute ^M^kM^faUk&wi. 2j. . linear in iJ. .^a.. . !_.Cv) can be solved in terms of and T>: 60 (77) Indeed by a completely algebraic (cohomological) method which is fully exhibited in Appendix A and B of reference [ 1 1 . Jb=o J o (74) >i. it is possible to show _3 that the first two consistency equations yield : (78) with (79) i e for some IIII • and I ._..
BLCCHI ET AL. Then the system °teft (^i*? ) up to order (77) is solved by the choice : T ^ . 0 and we want t o show t h a t t h e e q u a t i o n : (82) (83) can be solved in terms of »eff 0^) • Comparing with Eq.(69) we get : for some () _• and UL depending on the coefficients of V4 .0 (87) .XTinto the form : fields we can decompose £j.(69) we see that this is ensured if we can show that : fdx^V)=^of^ Af(x) for some IAQ linear in Ok • the dependence on the <p.259 2^6 C. Comparing Eq. ( X ) subject to the condition : o <S0(dx A ^ ( X ) .(79) with Eq. ^ [ ( < C ^ £ )(# jdx A^ (x) = O ( 6 8) from which tA and r e . i _y (84) Making explicit To exploit the constraint (82) we first write : <£ jd* Z£(X).<JU and K: = <T : (81) We are then left with the external field independent part of the anomaly A V .
the first cohomology equation for the gauge Lie algebra. Then applying Eq.(89). is. Here h (x) is the AdlerBardeen [ 1 j anomaly which has been discussed in Store's _6 lecture concerning the current algebra Ward identities 6 . <0 9> for some dimension four functional Zj ft . if the gauge group is semisimple : A ^*ibA + K>J . (91) . or. which is nothing but the WessZumino [17 consistency condition. Here £_± is a tensor symmetrical under the permutation of the first two indices. in other words. v ™ or ^Tc symbols and if the Lie algebra admits a non trivial inIf the variant completely symmetrical _L> AdlerBardeen anomaly is absent we get : ZC= ( < i x ( ^ ) ( x ) s 4 A c which completes the proof. tensor.GAUGK F I E L D MODF. )(x) (88) follow immediately. Such an anomaly can only arise if the tree lagrangian contains £^.(82) directly to A^} we get : and One can show that the general solution of Eq.LS 287 fah (c« *%M c ^) c^)>)=A P(c. (gcjc.
.288 4.E..B. We shall discuss here in some details the gauge independence of the physical S operator of the SU(2) H. First one should introduce the physical parameters (masses.(12)). is invariant under a group of global (non space dependent) rotations transforming the TT . 4 > j* . depends on five parameters..(4) and Eq. namely Q .K. K . through a system of normalization conditions. Then one has to specify a physical subspace within which the S operator is unitary and independent from the properties (masses.) of the ghost particles.(14). given in Eq. there remains to show that one can interpret it in physical terms in spite of the presence of many ghost fields.(12). " H . model whose classical limit has been discussed in the first section. BliCCHI ET AL. depends on five parameters. C . coupling constants) into the theory. Given a gauge model renormalizable in such a way as to preserve the Slavnov identity to all orders of perturbation theory.(14). CXu . and F is explicit in Eq. to the parameters which are left arbitrary in the lagrangian. The quantized fields are The classical lagrangian. namely : 2 a . These however have to be adjusted so that the coefficient of the term linear in . whose classical form is given in Eq. connecting them. GAUGE INVARIANCE C. Under this restriction the Slavnov transformations. e . C fields as vectors and leaving the O" field invariant.» ^ . The dependence on three of them. There are two more hidden parameters which can be introduced by the substitution : (92) c* * x ^* e > e /r The most general lagrangian invariant under Slavnov transformations : (93) L* 41 ) K U (compare with Eq.
=. One can alternatively specify the position of the poles in the transverse photon. ) and a coupling constant £ .>. owing to the global isotopic spin symmetry.o aTaT IP). all the ghost . one related with the & field vacuum expectation value. four specifying the transformation laws. the masses associated with the coupled longitudinal photon. as a consequence of the Slavnov invariance. the residues at these poles ( Z"A 2" K. by the normalization con ditions : (arm± f ' &\is>) X v ^ 4>.V). z a S*(f £•")[ ?<»'+ 0((p'»*f) • s s p (pshrO . Hence. r. five specifying the external field independent part of the lagrangian.TT channel are pairwise degenerated with those of the cc channel.' * ' K 'f ' 6 ' F leav ing free the unphysical parameters IT and % It is easy to show that. O" and cc propagators (m.^ (pv^) .GAUGI FIELD MODELS 289 O" vanishes (94) L 3! The theory thus depends on ten parameters. nUjp ). M..(94) which is equivalent to : (96) fix the values of z a ' z4 > f.*ft +O(IP"«^) These normalization conditions together with Eq.
290 C.(3. If ''CL'' is a parameter of our lagrangian we define the invariant derivative with respect to CL : It is easy to verify that : 3.1) = ^. <9 9> Da 2 (?.O. BICCHl LT AL. According to the analysis of section 3 it is possible to find an effective lagrangian fulfilling the Slavnow identity (20) and the normalization conditions (95) and (96) to all orders of perturbation theory.t)=° and (102) . masses are degenerate and fixed by the normalization conditions (95).(lOO) in the form : ^(At2).s 2 (3.V.(24)).•?)=£ (K*}(U) for some insertion (D. Here we shall sketch out the proof of Eq. we can write Eq.aza7)=0 and that : Since the Lowenstein action principles yields .(99) and Eq.) LJka of dimension four.(97) referring for the details to the existing literature fl3j. where is one of the parameters 7C. To prove in diis theory the gauge invariance of the physical S operator we have to show that: ^2f>hv. S^ 3*1*0 * X =0 ^ (97) (compare with Eq.
Now the set of the symmetrical insertions of our model is a linear space of dimension nine.5 (106) are such that the matrix with columns \>\tn* (107) i.GAUGl. We shall also call an insertion /inon physical if : Our aim is to prove that the symmetrical insertion .(102). one can construct explicitly (as in reference I 131 section 3C) in terms of functional differential operators four independent non physical insertions : AL =A> ••'** (105) L The remaining five independent symmetrical insertions : A50 <S) / A (S) L= V.s v L aTaTo(108) has non vanishing determinant : detii < n*° This can be easily proved by constructing explicitly the five insertions in the tree approximation. for a prescribed Slavnov operator "o ( P fixed) our theory depends on nine parameters.XslK. (Eq.(lOO)). YY)ekTT ) is n o n ZJ.FIELD MODI LS 29] In the following we shall call symmetrical an insertion satisfying Eq. since. Now the insertions (105) . Also.. (for physical.
.13j. Hence in particular : A" . .M . BIXCHl LT AL. lit CW C and + Z .(U2) . We shall first introduce a new coordinate frame for the unphysical degrees of freedom of our model. J (109) ^r(t^). taking into account Eq.292 C.. and (106) are a basis of the linear space of symmetrical insertions. ( o n) The independence from "X of the physical parameters Z A > £ is expressed by the system : m .^Am^)+i(Ay r)c^i).* Let us now give an outline of the connection between the Slavnov symmetry and the unitarity of the restriction of the S operator to the physical space 12. dAg A. I.(llO) is equivalent to : tj^'A'^o which completes the proof.. z . x '1TQ T A a T a T <r which. Namely we shall replace the longitudinal photon and TT fields by their independent linear combinations g* (the gauge function) and : .
r t (117) 0 Now the Slavnov identity can be translated into a symmetry property for S .. ^ j . Is is also easy to verify that the Fourier transformed gg propagator is independent from the momentum.:.S. 5 <r\ *. namely : S« Z2(J. reduction formulae [6j.G A U G i : I IlLD MODI LS 293 ~~ (114) where (115) o Owing to the Slavnov symmetry the gg propagator has _0 a. Introducing the complete system of asymptotic fields I*}. This greatly simplifies the analysis of the ghost contribution to the unitarity sum.l)j.Z. \ c h £jj and the corresponding wave operators K the scattering operator of our model is given by the L.0 (116) The main advantage of the restricted t'Hooft gauge is the absence of double poles in the propagators of the ghost fields. Indeed introducing the functional differential operator : . We shall also put ourselves in the restricted t'Hooft gauge defined by the normalization condition : '^fcitik*™ . _ a simple pole at P s ^<±>TT with the same residue as the cc propagator.
it turns out that the total contribution to the physical unitarity of states involving gg and (or) cc pairs vanish. owing to the anticommuting character of the ^>. BLCCHI 1 T AL.(25)) and the gg propagator is momentum independent.294 C.TT.  <s.U (X) (118) which corresponds to the field transformations (119) the Slavnov identity for the amputated Green functions on the mass shell simplifies to Z<S H«f. . fields. Since. This is the whole contribution coming from the unphysical states since the g field is decoupled from the physical states (Eq. their contribution to the unitarity sum is opposite in sign.D •= ° (120) The essential meaning of this equation is that on the mass shell the gg and cc pairs are equally coupled.
Concerning the possible extensions of our results.F. The extension to G. . If the gauge group is semisimple and all the fields are massive an algebraic analysis of the possible quantum corrections to the action principle shows that.GAUCiU F I I L D M 0 D 1 L S 295 CONCLUSION The gauge field models are characterized by the fulfillment of Slavnov identities. Lowenstein will show at the end of his lectures how the infrared problems connected with the gauge field models can be overcome at least in the extreme case (pure YangMills models) in which all the particules are massless [2. with non semisimple gauge group (i. indeed.M. the Slavnov identities can be fulfilled in the absence of the AdlerBardeen anomaly.18] . for example the Weinberg models) can be worked out supplementing the algebraic analysis of the anomalies by a power counting analysis similar to that used in reference 17J to deal with the invariantabelian anomalies in the lagrangian models with broken symmetry.e. We have also discussed in the SU(2) HiggsKibble model the connection between the Slavnov identity and the unitarity and gauge invariance of the physical S operator. The analysis of gauge independent local operators. should also be based on their Slavnov invariance l2. although not touched here.
LAM Nuovo Cimento 38.296 C. B. STORA Lectures given at this School.S. BECCHI. 9C n°1 (1973). Bonn. Phys. A complete bibliography about the theory of gauge fields can be found for instance in : E. A. J.M.S. D6. ZIMMERMANN Ann. Reports. LAM Phys. 2731 August 1973. M. 536570 (1973). P. LOWENSTEIN Lectures given at this School. Y. J. J. D. BREITENLOHNER. 82. VELTMAN Invited talk presented at the International Symposium on Electron and Photons at High Energies. W W. BREITENLOHNER Lectures given at chis School. BECCHI FT AL.734 (June 1975). ROUET. SCHWINGER Phys. R. LOWENSTEIN Lectures given at this School. 1754 (1965). 2145 (1972).P. MAISON Dimensional Renormalization and the Action Principle MPIPAE/PTh25/74 (May 1975). E. 918 (1951). R.W. LEE Phys. Rev. [7] C. ABERS. C. REFERENCES jl "] J. P. ZINNJUSTIN Lectures given at the International Summer Institute I 8j . STORA Renormalizable Theories with Symmetry Breaking Marseille Preprint 75/P. Rev. 77. SPEER Lectures given at this School.
BECCHI. 153 (1972).C. Cambridge Mass. [lo] L. R. J. B33. BECCHI. 1848 (1969). Marseille Preprint 75/P.N. A. BARDEEN Phys. 1156 (1966). Letters 52B. TAYLOR Nucl. Brandeis University Summer Institute in Theoretical Physics . ZUMINO Phys. i Mat. KIBBLE Phys. HIGGS Phys. A. A. ROUET. 42. SLAVNOV Teor. Phys. M [l7] 184. 29 (1957). 321 (1964). Letters 12. J.D.D M O D I as for Theoretical Physics. V.A. Letters. J. C. math. 25B. WESS.GAUGI: rni. Phys.B. T. 95 (1971). R. Rev. 154 (1961). GOLSTONE Nuovo Cimento [14] 19. STORA Phys.2 (the MIT Press. Letters 37B. Fiz. STORA Renormalization of Gauge Theories. C. Bonn 1974. ROUET. BROUT Phys. 132 (1964) . STORA Commun. 436 (1971). 1554 (1967). vol.W. [»] B. BECCHI. 297 F. Letters. Phys. [121 [ill 10. POPOV Phys. 145. 155. A. 127 (1975). Rev. [l3] C.723 (April 1975).A. .) W. 13. Rev. B. [9] P. 344 (1974). Rev. ROUET. R. R. ENGLERT. ZUMINO in Lectures on Elementary Particles and Quantum Field Theory. FADDEEV.
P„) = 0. the existence of such a fixed point. 6.271 VOLUME 30. 4 If we set P f = Atf °.9 The only exception would seem to be nonAbelian gauge theories. Recently.P d x%^D"yU')VM{g. y{gf)*0. one obtains naive scaling up to finite and calculable powers of InA.qi)exp[nJ0\{g(g. New Jersey 08540 (Received 27 April 1973) It is shown that a wide class of nonAbelian gauge theories have. We have found that they possess the remarkable feature.t))y{gf)}dt}. it has been a r gued for all but gauge theories.Pi) = XDr^\g{g. gf = 0. which hitherto have not been explored in this r e 1343 .e. As Wilson has stressed.. (n) (2) . Such asymptotically free theories will exhibit. the Euclidean momentum at which the theory is renormalized. (4) so that y(gf) is the anomalous dimension of the field. 2. perhaps unique among renormalizable theories.qi)^{nJ0\(g\g. D is the dimension (in mass units) of r solution of dg/dt= p(g).. It is suggested that Bjorken scaling may be obtained from stronginteraction dynamics based on nonAbelian gauge symmetry.7 In that case the anomalous dimensions of all operators vanish.t). We therefore suggest that one should look to a nonAbelian gauge theory of the strong interactions to provide the explanation for Bjorken scaling. might explain Bjorken scaling and the success of naive lightcone or partonmodel relations . g{g. of asymptotically a p proaching freefield theory. NonAbelian gauge theories have received much attention recently a s a means of constructing unified and renormalizable theories of the weak and electromagnetic interactions. and the structure of operator products at short distances is that of freefield theory. Princeton. If the physical coupling constant is in the domain of attraction of a UVstable fixed point.t'))dt']. Gross t and Frank Wilczek Joseph Henry Laboratories. we must a s sume the existence of a tower of operators with canonical dimensions. where q° a r e (nonexceptional) Euclidean momenta. The UV behavior of renormalizable field theories can be discussed using the renormalizationgroup equations. If we wish to explain Bjorken scaling. which has so far eluded fieldtheoretic understanding. Princeton University. Pig) and y{g) a r e finite functions of the renormalized coupling constant g. it appears that the fixed point at the origin. and g. for a theory of the strong interactions. the invariant coupling constant. Bjorken scaling. for matrix elements of currents between onmassshell states.0)=g. which is common to all theories. is the (3) where / = lnA. (n) The UV behavior of r (A — + <*>) is determined by the large/ behavior of g which in turn is controlled by the zeros of /3: P{gf) = Q. then (1) determines the A dependence of r ( n ) : rM^. (1) r asy is the asymptotic part of the oneparticleirreducible renormalized nparticle Green's function. 1 In this note we report on an investigation of the ultraviolet (UV) asymptotic behavior of such theories. the dimensions of operators at a fixed point a r e not canonical. Cn) ... that this can only occur if the fixed point is at the origin. the UV behavior is determined by the theory at the fixed point (g In general. then ^'n)is. 7 Therefore. freefieldtheory asymptotic behavior. and m is either the renormalized mass or.3 which for a theory involving one field (say g(p4) a r e [w3/8w+ P(g)d/3gny(g)]T^ISiPi. in the case of massless particles. These fixed points of the renormalizationgroup equations a r e said to be UV stable [infrared (IR) stable] iiggf as t+°° (•») for g{0) near gf. is not UV stable. Unfortunately. up to calculable logarithmic corrections. NUMBER 26 PHYSICAL REVIEW LETTERS 25 JUNE 1973 Ultraviolet Behavior of NonAbelian Gauge Theories* David J. 8.. so that the theory is asymptotically free. i.
T(R) = C2(R)d(R)/r. (10) where V(q>) contains cubic. (5) tion is given by LF= ifi(iy5 gyBaMa)ip+ mass terms.13 order. the renormalization is performed at an (arbitrary) Euclidean point. and one must deal with a new coupling constant. Let us consider a YangMills theory given by the Lagrangian F. and C2iR) is the quadratic Casimir operator of the representation. (9) where the Cabc a r e the structure constants of some (semisimple) Lie group G. Since the theory is massless. Zz and Z . a scalar meson transforming under a complex (real) r e p resentation R of the gauge group adds to /3 a term equal to ? (?) of Eq. a s a consequence of gauge invariance and the KallenLehman r e p r e sentation. Consider the Lagrangian for the coupling of scalars belonging to a representation R of G: £ = i[(auigBll'M°)y]2\(yv)2+V(v). NonAbelian theories have no such requirement. g*T. Although the fermions tend to destabilize the origin. The problem with scalar mesons is that they necessarily have their own quartic couplings. Their contribution to P(g) is small. The only effect of the fermions is to change the value of Pig) by the amount11 Pr(R) = (g3/16tl)^T{R). (6) (For a thorough discussion of the renormalization see the work of Lee and ZinnJustin. etc. where M" a r e the matrices of some r e p resentation R of the gauge group G.e. additional PY=(g3/lG*2)1iC2(G) + 0(g% (8) quartic terms invariant under G. in the case of SU(3): / 3 V = . could have any sign at g = 0. ^ / 6 ( ^ w = (^+^)gK>. however. quadratic. the number of generators. It is easy to indT(T. (9). will remain m a s s less until the gauge symmetry is spontaneously broken. the wavefunction renormalization constant Z3(g. Thus Pig). The renormalizationgroup equations have an additional term. One might hope that this would be a consequence of the dynamics.272 VOLUME 30. perhaps. The vector mesons. A/TO) will be defined in terms of the unrenormalized vectormeson propagator D^^ „(in the Landau gauge). 14 but at the present the only known way of achieving this is to introduce scalar Higgs mesons. For example. whereas 0^(3) =f. so that one could accomodate as many as sixteen triplets. NUMBER 26 PHYSICAL REVIEW LETTERS 25 JUNE 1973 gard. and thus P(g)Sig3 which leads to IR stability at g = 0.g2)/dt=g2[AT2 + BT+C] (11) corporate fermions into such a theory without d e 4 (where we have neglected t e r m s of o r d e r s . where C2{G) is the quadratic Casimir operator of 0 xig> *•) 3 / 9 A. and g — 0 as / — <x> a s long a s the physical coupling con. In order to investigate the stability of the origin. there is room to spare. The fermion interac 1344 . and Zx the chargerenormalization constant. One can therefore construct many asymptotically free theories with fermions. and we find that 12 where Tr(iirjW*) = T(R)6ab.g.. The soluboth g and A [if there a r e other quartic invariants tion of (3) is then g2(t)=£2/(l 2^rg'H). 11 In the Landau gauge they a r e identical with (1).v°=SllAu°8vAll'gCabcA„l'Av°.. stroying the UV stability. a r e gauge dependent and can be greater than 1. it is sufficient to calculate 3 to lowest order in perturbation theory. £^8) = 4. whose nonvanishing vacuum expectation values break the symmetry. a n d o n e must investigate the UV the adjoint representation of the group G: Yjbccabc stability of the origin (g= A = 0) with respect to xcdbc = C2(G)5ai[e. In Abelian gauge theories Z3 = Zx = 1 g2C InA (C>0). i. which must be gauge independent in lowest order. d(R) is the dimension of the representation R. The structure of the renormalizastant g is in the domain of attraction of the o r i tiongroup equation for g is unchanged to lowest gin. and r is the order of the group. We have calculated Zx and Z 3 for the above Lagrangian. The introduction of scalar mesons has a very destabilizing effect on the UV stability of the o r i gin.. whereas for the coupling constant I = \/g2 We have thus established that for all nonAbel we have11 ian gauge theories based on semisimple Lie groups the origin is UV stable. 10 ) The r e normalizationgroup equations for this theory a r e easily derived. and linear terms in <p (which have no effect on the UV b e havior of the theory) plus. C2(SU(A0)=N]. To this order we have where A is a UV cutoff.in V(cp) there will be additional coupling constants to consider]. For example.1 1 . t.
N) r e p r e s e n t a t i o n for N^ 5. ^ . ZinnJustin. In a l l of t h e s e m o d e l s i t i s n e c e s s a r y for the t h e o r y to contain a l a r g e n u m b e r of f e r m i o n s in o r d e r to m a k e j3 f s m a l l . and many o t h e r s . in none of t h e s e m o d e l s can t h e gauge s y m m e t r y be totally b r o k e n by the Higgs m e c h a n i s m . Unfortunately. O n e p a r t i c u l a r l y a p p e a l i n g model i s b a s e d on t h r e e t r i p l e t s 1 6 of f e r m i o n s . We have found many e x a m p l e s of t h e o r i e s which contain s c a l a r m e s o n s and a r e UV s t a b l e . Gross. 3 C. to be published. Phys. to be published.M a n n ' s SU(3) ®SU(3) a s a global s y m m e t r y a n d an SU(3) " c o l o r " gauge group to p r o v i d e t h e s t r o n g i n t e r a c t i o n s . Wilczek. t h e g e n e r a t o r s of t h e s t r o n g i n t e r a c t i o n gauge g r o u p c o m m u t e with o r d i n a r y SU(3) <g>SU(3) c u r r e n t s and mix q u a r k s with the s a m e i s o s p i n and h y p e r c h a r g e but different " c o l o r . Gross and F . W. 8 A. Phys. 95. 13 K.273 VOLUME 30. Rev. 1972 (to be published). Rev. 4 The basic assumption underlying the derivation and utilization of the renormalization group equations is that the large Euclidean momentum behavior of the theory is the same a s the sum._18. 1 7 will be a s y m p t o t i c a l l y free if we a g a i n i g n o r e t h e c o m p l i c a t i o n s due to the H i g g s p a r t i c l e s . Rev. W e i n b e r g . Wilson has suggested that the coupling constants of the strong interactions a r e determined to be LRstable fixed points. 1 4 T h i s i s s u g g e s t e d by t h e IR i n s t a b i l i t y of the t h e o r i e s . We have i n v e s t i g a t e d t h e s t r u c t u r e of t h e s e e q u a t i o n s for a l a r g e c l a s s of gauge t h e o r i e s and r e p r e s e n t a t i o n s of t h e s c a l a r m e s o n s . D. at least in a domain about the origin. W. see B.s t r u c t u r e constant a s the value of t h e U V . Rev. NUMBER 26 PHYSICAL REVIEW LETTERS 25 JUNE 1973 g4T2. which a s s u r e s u s that p e r t u r b a t i o n t h e o r y i s not t r u s t w o r t h y with r e s p e c t t o the s t a b i l i t y of the s y m m e t r i c t h e o r y n o r to i t s p a r t i c l e content. It t h u s a p p e a r s to b e v e r y difficult t o r e t a i n UV s t a b i l i t y and b r e a k t h e gauge s y m m e t r y by e x p l i c i t l y i n t r o d u c i n g Higgs m e s o n s . UV instability i n evitably o c c u r s . 5 K. Politzer [private communication.h a n d side of (11) h a s p o s i t i v e z e r o s (C >0. we would p r e fer to b e l i e v e t h a t s p o n t a n e o u s s y m m e t r y b r e a k ing would a r i s e d y n a m i c a l l y . Math. 1346 (1973)]. Since t h e H i g g s m e s o n s a r e s o r e s t r i c t i v e ._19. National Accelerator Laboratory. Rev. G. D 5. T h e r e q u i r e m e n t that t h e i n t e r a c t i o n s of t h e s c a l a r m e s o n s b e r e n o r m a l i z a b l e s o s e v e r e l y c o n s t r a i n s t h e f o r m of L a g r a n g i a n that t h e ground s t a t e i n v a r i a b l y i s i n v a r i a n t u n d e r s o m e n o n . 30.+ 0 a s t — <*>. T C. T h a t i s . 1 1 T h e s e include (a) SU(iV) if the s c a l a r m e s o n s belong to the adjoint r e p r e s e n t a t i o n for N> 6. B<0. Phys. to all o r d e r s . t0 B. F4462071C0180. E. Phys. Batavia. Symanzik. Illinois. G G. 3121 (1972). For an extensive review as well as a list of references.A b e l i a n subgroup of the gauge g r o u p . Phys. built on s e m i s i m p l e L i e g r o u p s . If t h e r i g h t . Callan. J. Gross. o t h e r w i s e J 1 a p p r o a c h e s z e r o too r a p i d l y for the v e c t o r m e s o n s to s t a b i l i z e t h e s c a l a r couplings. 1541 (1970). to be published. to be published. With t h i s hope in mind one can c o n s t r u c t many i n t e r e s t i n g m o d e l s of the s t r o n g i n t e r a c t i o n s . s i n c e A i s s t r i c t l y p o s i t i v e and A m u s t b e p o s i t i v e . Commun. "Full details will be given in a forthcoming publication: D. Zee. Phys. Rev. 1818 (1971). Callan and D. D 3 . Lee. (c) SV(N) with t h e s c a l a r s t r a n s f o r m i n g a s a s y m m e t r i c t e n s o r for N * 9. For nongauge theories the IR stability of the origin in fourdimensional field theories implies that theories so constructed a r e trivial. then for T l e s s than the l a r g e r z e r o of (11) w e will have t h a t A . 227 (1970). Low. 1 5 T h e v e c t o r m e s o n s c o n t r i b u t e t o B and C a n d tend t o s t a b i l i z e t h e o r i g i n . and t h e s t r u c t u r e of the o p e r a t o r p r o d u c t e x p a n s i o n of e l e c t r o m a g n e t i c o r weak c u r r e n t s i s ( a s s u m i n g that t h e s t r o n g coupling constant i s in the d o m a i n of a t t r a c t i o n of t h e origin!) e s s e n t i a l l y that of the free q u a r k model (up to c a l c u l a b l e logarithmic corrections). 2 M. Lett. with G e l l . and following Letter. Phys. 9 S. G. •Research supported by the U. Johnson. Sloan Foundation Research Fellow. Air Force Office of Scientific Research under Contract No. Our results suggest that nonAbelian gauge theories might possess IRstable fixed points at nonvanishing values of the coupling constants. GellMann and F . In t h e a b s e n c e of v e c t o r m e s o n s (g=0) t h i s equation i s UV unstable a t A=0. 1345 .S. Lee and J. we note that t h e o r i e s of the weak and e l e c t r o m a g n e t i c i n t e r a c t i o n s . K. Willey.11 F i n a l l y . in Proceedings of the Sixteenth International Conference on High Energy Physics. Coleman and D. Wilson. B 2. " In such a model the v e c t o r m e s o n s a r e n e u t r a l . a n d £ 4 r 3 ) . T h i s s u g g e s t s that t h e p r o g r a m of B a k e r . 12 After completion of this calculation we were informed of an independent calculation of /3 for gauge theories coupled to fermions by H. Lett.s t a b l e fixed point in quantum e l e c t r o d y n a m i c s might fail f o r s u c h t h e o r i e s . J . If one t r i e s t o o v e r c o m e t h i s by l a r g e r r e p r e s e n t a t i o n s for t h e s c a l a r m e s o n s . 1300 (1954). of the leading powers in perturbation theory. a n d B2 > 4AC). tAlfred P. and Adler 1 8 t o c a l c u l a t e t h e f i n e . 1264(1967). (b) S\J(N) <g>SU(iV) if t h e s c a l a r s b e l o n g t o t h e (N. to be published. J . Parisi.
Illinois. R e n o r m a l i z a t i o n of QED m u s t b e c a r r i e d out a t o f f . But specific m o d e l s in four d i m e n s i o n s that had been i n v e s t i g a t e d yielded (in t h i s context) d i s appointing r e s u l t s . Cambridge. Batavia. Under the hypothesis that spontaneous symmetry breakdown is of dynamical origin. D 5. T h e effective coupling goes to z e r o for l a r g e m o m e n t a . Thus. CERN Report No. Symanzik (to be published) has recently suggested that one consider a Xtp" theory with a negative A to achieve UV stability at A=0. that in such theory the groundstate energy is unbounded from below (S. F o r s m a l l e2. Georgi and S. T h e r e n o r m a l i z a t i o n g r o u p technique shows that t h e effective coupling g r o w s with m o menta. 345 (1961). and hence a c a l c u lation to any finite o r d e r i s m e a n i n g l e s s in t h i s d o m a i n . and M. Weinberg. W.s h e l l points b e c a u s e of i n f r a r e d d i v e r g e n c e s . see S. However. Phys.m a s s .M i l l s t h e o r y . 1888 (1973).g r o u p t e c h n i q u e s hold g r e a t p r o m i s e for studying s h o r t . L.S y m a n z i k equations c o m monly called fi(g) i s negative n e a r g=0. NUMBER 26 U PHYSICAL REVIEW 16 LETTERS 25 J U N E 1973 Y. (Received 3 May 1973) Jefferson Physical laboratories. Glashow. R e n o r m a l i z a t i o n .d i s t a n c e and s t r o n g coupling p r o b l e m s in field t h e o r y . Phys. 1494 (1972). 17 H. Bardeen. whose coupling could be strong. Phys. t h e r e is no s e n s e in which the i n t e r a c t ing t h e o r y i s a s m a l l p e r t u r b a t i o n on a f r e e m u l tiplet of m a s s l e s s v e c t o r m e s o n s . though one can well a p p r o x i m a t e a s y m p t o t i c G r e e n ' s functions. F o r t h e s e o n e . David P o l i t z e r Harvard University. Coleman. Massachusetts 02138 An explicit calculation shows perturbation theory to be arbitrarily good for the deep Euclidean Green's functions of any YangMills theory and of many YangMills theories with fermions. The t r u l y catastrophic infrared problem makes a symmetric particle interpretation impossible. S."+. The r e n o r m a l i z a t i o n . in Proceedings of the Sixteenth International Conference on High Energy Physics. Nambu and G. t h e m a s s . It i s c o m m o n l y said that for m o m e n t a such t h a t e2ln(p2/m2) ~ 1..o r d e r c o r r e c t i o n s b e c o m e c o m p a r a b l e . t o what p a r t i c l e s t a t e s do they r e f e r ? C o n s i d e r t h e o r i e s defined by the L a g r a n g i a n £ ={f„„'F. 3 T h i s note r e p o r t s an i n t r i g u i n g c o n t r a r y finding for any g e n e r a l i z e d Y a n g . The c o n s t r a s t with quantum e l e c t r o d y n a m i c s (QED) might b e illuminating. JonaLasino. A. But the l a r g e p2 d i v e r g e n c e r e p r e s e n t s a r e a l b r e a k d o w n of p e r t u r b a t i o n t h e o r y .c o u p l i n g . Weinberg. Rev. but a s p2's a p p r o a c h z e r o .g r o u p technique 4 p r o v i d e s a s o m e what opaque a n a l y s i s of t h i s situation. Adler. But what about the inevitable l o g a r i t h m s of m o m e n t a t h a t grow a s w e a p p r o a c h the m a s s shell or a s s o m e m o m e n t a go to infinity ? In QED. 1972 (to be published). Even for a r b i t r a r i l y s m a l l g2. S. Rev. Fritzsch. we expect p e r t u r b a t i o n t h e o r y to b e good in s o m e neighborhood of the n o r m a l i z a t i o n point. Loosely speaking. Rev. Thus p e r t u r b a t i o n t h e o r y t e l l s nothing about t h e m a s s .274 VOLUME 30.1irfltJ*J1 where (1) . 1 ' 2 Symanzik 2 h a s e m p h a s i z e d t h e r o l e that p e r t u r b a t i o n t h e o r y might play in a p p r o x i m a t i n g t h e o t h e r w i s e u n known functions that o c c u r in t h e s e d i s c u s s i o n s . J28. Rev. Coleman and E. these symmetric Green's functions are the asymptotic forms of the physically significant spontaneously broken solution. c o m p e n s a t i n g for t h e fact t h a t t h e r e a r e m o r e and m o r e of t h e m . Reliable Perturbative Results for Strong Interactions?* H. GellMann. using the renormalizationgroup equations. 1962 (1972).c o n s t a n t t h e o r i e s (or g e n e r a l i z a tions involving p r o d u c t groups) t h e coefficient function in t h e C a l l a n . 15 K. 18 For a review of this program. CERNTH1538. Lett. when we take due account of t h e e x p e r i m e n t a l s i t u a t i o n . Phys. T h e b e h a v i o r in t h e two m o m e n t u m r e g i m e s i s r e v e r s e d in a Y a n g . L. D 2 . 5 t h e effective coupling of soft photons 1346 goes to z e r o . National Accelerator Laboratory. 1972 (to be published) .s h e l l d i v e r g e n c e s do not o c c u r in o b s e r v a b l e p r e d i c t i o n s ._122. one can show.M i l l s t h e o r y and t h e o r i e s including a wide c l a s s of f e r m i o n r e p r e s e n t a t i o n s .s h e l l s t r u c t u r e of the s y m m e t r i c t h e o r y . H. private communication). h i g h e r . h i g h e r o r d e r s b e c o m e c o m p a r a b l e .
g ) = 1 (up to a phase convention. To first order. n ' . the contribution to the fermionvector threepoint vertex correction proportional to the firstorder QED correction needs no subtractions and contains no reference to M.t). . it r e mains zero under finite renormalizations. But if the fields are rescaled as in Eq.t) is defined by 9g'/M=li(g').) The normalizations of the conventionally defined irreducible vertices for n mesons and n' fermions. involving a momentumscaledependent effective couplingg'(g. that to this order the coupling constants of product groups do not enter into each other's (9 functions.c 1 . (3) /„<*/><* = 2c 1 5 al . For the case where there a r e no fermions.VOLUME 30. (4) It is also apparent. it can be simulated by introducing two multiplets of spinor fields with the same group transformations but opposite statistics. which vanishes in the Landau gauge. Hence a should occur in Eq. The improved vertex functions are con structed from the straightforward perturbative ones.f c 2 ) ( ^ / 4 1 r ) 2 + 0 ( ^ ) . [For SU(2). and one would have to study the r n .p J ^ T j / y J l ^Cl(g/^)2ln(p2/M2)}. 1347 . Consider gauges defined by a in the zerothorder propagator . ln(p2/M2)]. where t = iln(s/M 2 ) and s sets the scale. Calculation of the remaining correction.) Alternatively.n unchanged. one can study the corrections to the threemeson vertex. . the meson inverse propagator is J W where P) = 0'\g„ „p* +/»„/» J[l + (AK . cl=l. involving the meson selfcoupling. (5) The normalization condition is F(l. by inspecting the graphs. P. which had been used implicitly to define^. r for arbitrary a to determine the coefficient functions perturbatively.§c2)g(g/W+ 0(gs). 0(*) = . e. (2) much as g does. (2).f c2){g/^f tr(TaTb) = 2c2Sab. But for a =0 initially. is no longer p r e s ent. Applying Eq. Similarly. NUMBER 26 PHYSICAL REVIEW LETTERS 25 JUNE 1973 and Dlf=a"6IJigA"'TtJ: t h e / ° ' c a r e the group structure constants. c 2 (isodoublet) = \. a must be changed to leave T2 invariant. (2) to these functions at their normalization points yields y#(*) = 0 + O(**). The generalized Ward identities 7 imply that there are no higherorder corrections to the longitudinal part.( f c. g'(g.<P*) = gu» + PuPv/p2 ^ nPuPv ^s)~^nyA(g) + n % t e ) ) l " . Define F by r\„„ 3 a i c (P. and cx>0 and c 2 > 0 . (2) Putting it in this form makes use of the first available simplification. and c 2 (isotriplet) = 1. proper choice of gauge. must refer to some mass M..0)=g. / • " ( O ^ . The physical effects of internal fermions are canceled by the ghosts —spinor fields with Bose statistics.g. the coefficient functions can be obtained by setting c 2 = 0.0) = / ° 6 ' ( p x g l l v +P]lgvX 2  2pvgXll)gF(p2/M2. and the T" a r e representation matrices corresponding to the fermion multiplet. (7) g'(g.] To first order (only)." ' = 0.g2). y x (^) = ( f . r n . s=J^(p2). so it suffices to study the theory in a Landau gauge.) To first order F^l + Zc^gMirflnip^/M ) 2 (6) which yields the same /S as described above. the fermion selfenergy is proportional to the selfenergy in massless QED. The renormalizationgroup equation reads {M^ + Equation (2) describes how finite renormalizations accompanied by a change ing and a rescaling of the fields leave the Tn. (Even though the fermionvector vertex. (One may be interested in models with massless fermions because of their group structure or because they have the same asymptotic forms 6 as massive theories. The renormalizationgroup "improved" perturbation theory4*5 extends results valid near the normalization point by effectively moving that point. yields ^ .
Particular models which are ultraviolet stable as well as spontaneously asymmetric have been found. (1) [withjr less than the first zero of 0(g)] to the model with^ arbitrarily small by a change in mass scale. 3 infrared instability is a necessary condition for spontaneous symmetry breakdown. DESY Report No. 11 In the renormalizable massless theories including scalars that have been studied. the problem with the NambuJonaLasinio model is not its nonrenormalizability but that in the domain of large g0A2. But in m a s s less scalar QED. the fields obtain (in general massive) particle interpretation—the Higgs phenomenon. It is worth remembering that successive orders of perturbation theory give the behavior of /3 for infinitesimal £• and. Adler. But the condition that the oneloop approximation imply vacuum degeneracy requires that the expansion parameter be large. DESY 72/73. /3= . S. Making a polynomial fit to a perturbative result for 3 is pure conjecture. we make the following hypothesis: The gauge symmetry breaks down spontaneously as a result of the dynamics. which one can do because of the cutoff. F4462070C0030. As yet. Consequently. (In the framework of the original solution. nothing is known about the particle spectrum. Air Force Office of Scientific Research under Contract No. "Some Speculations on High Energy Quantum Electrodynamics" (to be published). Rev. "The Renormalization Group and the € Expansion" (to be published). 2 K. S. 9 But these theories are not in general ultraviolet stable in terms of the additional coupling constants that must be introduced. where spontaneous breakdown is alleged to occur. A study of the Green's functions in the oneloop approximation yields all the original results. They all share the same asymptotic Green's functions. But the existence of the vacuumdegenerate solution requires the dimensionless parameter characterizing the theory to satisfy g0A2>2ir2. Xtp* is stable for small \ because the oneloop corrections are small. rendering the application of perturbation theory suspect. 9 The model is defined by & = uy H + gn[{^d)2 . and the latter especially for his help in the computations. Dj>. higher loop corrections are likely to dominate. Phys. including the possibility of a dynamical determination of the dimensionless coupling constant. The T"'"'show a welldefined slow approach to quasifree field values. Hypothesizing that 8 stays negative (at least into the domain of strong coupling constant) r e lates all theories defined by Eq. Baker. To utilize this result. K. 2 a berg. strictly speaking. Johnson and M.276 VOLUME 30.(Jy^Y\ (9) (8) Thus for a pure meson theory or for theories including not too many fermions (in the sense that c 2<"T c i)> S' goes to zero for asymptotic momenta. i. But the CallanSymanzik analysis says that the asymptotic Green's functions for the "dressed" fundamental fermion and vector fields are the symmetric functions discussed above. 'Of central importance is the work reviewed in K.e. t~<*>. Wilson and J. say nothing about finite g. But this is the condition that the oneloop correction to fermionfermion scattering be at least as important as the tree approximation. 3021 (1972). in terms of which the group transformation properties of the vacuum can be stud ied.bg3. or particles describable only by composite fields. g' ^g'/(l+2bg t).) In theories defined by Eq. The requirement is just that \<e4. 8 [An alternative is to introduce fundamental scalar fields. photonloop corrections of order e 4 can dominate over the lowest order <pcp scattering (order X) for both A and e arbitrarily small. the lowenergy dynamics.. 1972 . *Work supported in part by the U.10 But gauge theories of fermions (only) have aesthetic attractions. Symanzik. Define a scalar <p(x)=gaW)Hx) and an analogous pseudoscalar. who have offered insights and advice freely. The situation is similar in the renormalizable models. NUMBER 26 PHYSICAL RE IE W L E T T E R S 25 JUNE 1973 For the approximate /3's derived above. differing only by how large is asymptotic. 9 ] Hypotheses of this type go back to the work of Nambu and collaborators. (1). more complex infinite chains and selfenergy graphs dominate over the ones studied. The author thanks Sidney Coleman and Erick Weinberg. 12 The model of Nambu and JonaLasinio can be treated by the methods of Coleman and Wein1348 and the stipulation that the momentum integrals are cut off at some Euclidean mass squared A2. composite scalar densities can also be defined and studied in perturbation theory. Kogut. In this light.
30. in the Proceedings of the 1971 International Summer School "Ettore Majorana" (Academic. JonaLasinio. 1349 . Lee and J. "Study of the Renormalization Group for Small Coupling Constants" (to be published). to be published). S. In particular. Shirkov. 2). NUMBER 26 PHYSICAL REVIEW LETTERS 25 JUNE 1973 (to be published). but for the general theory and derivations see Refs. 2. Coleman and E. 9 Coleman and Weinberg. the potential whose minimum determines the vacuum decreases without bound for large field. 5. 3 A. Weinberg. 2) and hence infrared unstable but cannot be physically interpreted in perturbation theory. Wilczek. 12 A<?4 theory with A < 0 is ultraviolet stable (Ref. B33. 9.. Phys. 3121 (1972). and 4. 7 E. 'Definitions of the relevant quantities will be given. Rev. e Asymptotic refers to a particular set of Euclidean momenta as they a r e collectively scaled upward. W. N. Using the computations of Ref. Nambu and G. Gross and F. etc. 1959). whose conventions we follow. Coleman. 1888 (1973). S. 173 (1971). New York. 4 N. Phys. 345 (1961). ZinnJustin. preceding Letter [Phys. 1. for A<0 "improved" perturbation Iheory is arbitrarily good for large field strengths. and references therein. Phys. Rev.277 VOLUME 30. n Y . g. which also include details of Feynman rules. New York. D. Rev. V._122. Bogoliubov and D. B. Phys. 'Configurations where the symmetric theory has infrared singularities not present in the massive case a r e discussed in detail by Symanzik (Ref. 't Hooft. D 5. Ref. Rev. Nucl. 1343 (1973)]. Introduction to the Theory of Quantized Fields (Interscience. Lett. D 1_. Zee. G. regularization.
the solution of which problems will presumably require a more drastic revision of our fundamental concepts than any that have gone before. 1931.. The most powerful method of advance that can be suggested at present is to employ all the resources of pure mathematics in attempts to perfect and generalise the mathematical formalism that forms the existing basis of theoretical physics. The stead}r progress of physics requires for its theoretical formulation a mathematics that gets continually more advanced. Quite likely these changes will be so great that it will be beyond the power of human intelligence to get the necessary new ideas by direct attempts to formulate the experimental data in mathematical terms. e. to try to interpret the new mathematical features in terms of physical entities (by a process like Eddington's Principle of Identification). What.60 Quantised Singularities in the Electromagnetic Field. It seems likely that this process of increasing abstraction will continue in the future and that advance in physics is to be associated with a continual modification and generalisation of the axioms at the base of the mathematics rather than with a logical development of any one mathematical scheme on a fixed foundation. The theoretical worker in the future will therefore have to proceed in a more indirect way. however. (Received May 29. Cambridge. .) § 1. have now been found to be very necessar}^ for the description of general facts of the physical world. but would rest on a permanent basis of axioms and definitions. while actually the modern physical developments have required a mathematics that continually shifts its foundations and gets more abstract. There are at present fundamental problems in theoretical physics awaiting solution. Noneuclidean geometry and noncommutative algebra.R. F. By P.g. the relativistic formulation of quantum mechanics and the nature of atomic nuclei (to be followed by more difficult ones such as the problem of life).S. which were at one time considered to be purely fictions of the mind and pastimes for logical thinkers. namely. M. This is only natural and to be expected. John's College. it was expected that the mathematics would get more and more complicated. and after each success in this direction.. A. DIRAC. Introduction. was not expected by the scientific workers of the last century was the particular form that the line of advancement of the mathematics would take. St.
. 361 (1930). Weyl. P. but if they could be produced experimentally in high vacuum they would be quite stable and amenable to observation. if it collides with an electron.Tamm. We may call such a particle an antielectron. Camb. Soc.. Oppenheimer. 61 A recent paper by the author* may possibly be regarded as a small step according to this general scheme of advance. having the same mass and opposite charge to an electron. p. 360 (1930). being something exceptional. Subsequent investigations.' vol. Physik. of the negativeenergy states for electrons are occupied.. ' Pays. ' Phys. by saying that in the physical world almost all the negativeenergy states are already occupied. ' Z.' A.' vol. Soc. unknown to experimental physics. 35. 126. so that our ordinary electrons of positive energy cannot fall into them. Rev. Roy.' 2nd ed. § J. which on this view really exist. all. 234 (1931). Rev. the two will have a chance of annihilating one another much too great to be consistent with the known stability of matter. p. vol. and not merely nearly all. The question then arises as to the physical interpretation of the negativeenergy states. J.j It thus appears that we must abandon the identification of the holes with protons and must find some other interpretation for them. Oppenheimer. It was proposed to get over this difficulty. 26. t H. R. ' Proc.§ we can assume that in the world as we know it. 35. 545 (1930) . 11. Philos. would be a new kind of particle.' vol. making use of Pauli's Exclusion Principle which does not allow more than one electron in any state. Dirao. The mathematical formalism at that time involved a serious difficulty through its prediction of negative kinetic energy values for an electron.' vol. Following Oppenheimer. however. p. We should not expect to find any of them in nature. but an unoccupied one of these states. p. p. It was shown that one of these holes would appear to us as a particle with a positive energy and a positive charge and it was suggested that this particle should be identified with a proton. 62. We should expect the uniformly filled distribution of negativeenergy states to be completely unobservable to us. An encounter between two hard yrays (of energy at least half a million volts) could lead to the creation simultaneously of an electron and antielectron. the probability of occurrence of this process being of the same order of magnitude as that of the collision of the two yrays on the assumption that they are spheres of the same size as classical * ' Proc. p.. have shown that this particle necessarily has the same mass as an electronf and also that. if there were one. R.. on account of their rapid rate of recombination with electrons. A hole. ' Gruppentheorie und Quantenmechanik. should make its presence felt as a kind of hole. 562 (1930). 939 (1930).Quantised Singularities in Electromagnetic Field.
It shows. y. analogous to the fact that the symmetry between electrons and protons is not forced when we adopt Oppenheimer's interpretation. while it looks at first as though it will give a theoretical value for e. We express <\> in the form <\> = My. . Dirac. We consider a particle whose motion is represented by a wave function ^. all of which normally are occupied. M. I t does not. Theor}T at present is quite unable to suggest a reason why there should be any differences between electrons and protons. electrons. a symmetry between electricity and magnetism quite foreign to current views. Nonintegrable Phases for Wave Functions. The precise form of the wave equation and whether it is relativistic or not. This probability is negligible. which is a function of x. an unoccupied one appearing as an antiproton. This smallest charge is known to exist experimentally and to have the value e given approximately by* Ac/e2 = 137. For a given state of motion of the particle. are not important for the present theory. denoting the amplitude and phase of the wave function.62 P. but with the reason for the existence of a smallest electric charge. however. which must be of modulus unity if we impose the condition that <]> shall be normalised. z and t. Without this symmetry. § 2. it introduces quantitative differences between electricity and magnetism so large that one can understand why their qualitative similarities have not been discovered experimentally up to the present. not with electrons and protons. with the intensities of yrays at present available. completely undetermined and if we insert the experimental value 137 in our theory. the ratio on the lefthand side of (1) remains. * h means Planck's constant divided by 27t. in fact. Presumably the protons will have their own negativeenergy states. y. The object of the present paper is to put forward a new idea which is in many respects comparable with this one about negative energies. <\i will be determined except for an arbitrary constant numerical coefficient. (1) The theory of this paper. force a complete symmetry. The protons on the above view are quite unconnected with electrons. (2) where A and y are real functions of x. It will be concerned essentially. is found when worked out to give a connection between the smallest electric charge and the smallest magnetic pole. A. from the theoretical standpoint. however. z and t.
but only a definite difference in values for any two points.Quantised Singularities in Electromagnetic Field. For two distant points there will then be a definite phase difference only relative to some curve joining them and different curves will in general give different phase differences. We may go further and assume that this difference is not definite unless the two points are neighbouring. namely. The integral \<f>mi\indxdydz is a number. we may have to make use of the product j>mijn. Whenever a tyn appears. however. which must also be considered. We may assume that y has no definite value at a particular point. whether neighbouring or not. This density is independent of the phase of the wave function. Thus the value of y at a particular point has no physical meaning and only the difference between the values of y at two different points is of any importance. This immediately suggests a generalisation of the formalism. when extended so as to give the same uncertainty of phase for transformation functions and matrices representing observables (referring to representations in which x. must have a definite phase difference between any two points. it will at . The total change in phase when one goes round a closed curve need not vanish. the square of whose modulus has a physical meaning. We thus get the general result:— The change in phase of a wave function round any closed curve must be the same f or all the wave functions. This requires that the change in phase in <\in round a closed curve shall be equal and opposite to that in 4>m and hence the same as that in <\im. is sufficient to insure that the nonmtegrability of phase gives rise to no ambiguity in all applications of the theory. In order that the integral may have a definite modulus the integrand. so that no trouble will be caused in this connection by any indeterminacy of phase. 63 The mdeterrninacy in >h then consists in the possible addition of an arbitrary constant to the phase y. Let us examine the conditions necessary for this nonintegrability of phase not to give rise to ambiguity in the applications of the theory. if it is not multiplied into a <j>m. although it need not have a definite phase at each point. If we take two different wave functions <bm and <\>n. the probability of agreement of the two states. which has a direct physical meaning. Thus the change in phase in 4>m^n round a closed curve must vanish. y and z are diagonal) as for wave functions. If we multiply tjj by its conjugate complex j> we get the density function. It can easily be seen that this condition. There are other more general kinds of applications.
64 P. representing an observable a. any rate be multiplied into something of a similar nature to a <f>m. z. This transformation function will have the same uncertainty of phase as a <f>. z". except for a constant which does not matter. Dirac. The superposition principle for wave functions will be discussed a little later and when this point is settled it will complete the proof that all the general operations of quantum mechanics can be carried through exactty as though there were no uncertainty in the phase at all. the observables \ are diagonal. For example. This requires that [3 shall not be a function of x.<\>n. y. it must be multiplied by the transformation function (£  xyzt) and integrated with respect to x. M. as a product + = +i« ip . which will result in the uncertainty of phase cancelling out. (3) where <\i1 is any ordinary wave function (i. the uncertainty in the phase as concerns the column [specified by x". if <\in is to be transformed to another representation in which./c. For the mathematical treatment of the question we express t^.  (*.. so that the transformed wave function will have its phase determinate. Again. y". which do not in general satisfy the conditions of integrability dKx/dy = dKy/dx.e. by Stokes' theorem. y and z. t having a definite value at each point. The above result that the change in phase round a closed curve must be the same for all wave functions means that this change in phase must be something determined by the dynamical system itself (and perhaps also partly by the representation) and must be independent of which state of the system is considered. it appears that the nonintegrability of phase must be connected with the field of force in which the particle moves. The uncertainty of phase is thus put in the factor eI/J. one with a definite phase at each point) whose modulus is everywhere equal to the modulus of ty. say. except for a constant independent of £. The change in phase round a closed curve will now be. etc. if we multiply <\>n by a matrix (x'y'z't  a  x"y"z"t). dS). ds) =  (curl. A. (4) . As our dynamical system is merely a simple particle. t] will cancel the uncertainty in <]in and the uncertainty as concerns the row will survive and give the necessary uncertainty in the new wave function a. more generally than (2). but ( must have B definite derivatives _ as K _ ap Ky _ ap K _ ap K '~dx' ~dy' °~dz' °~dt' at each point.
since ^ is just an ordinary wave function with a definite phase. Let ua assume that <\i satisfies the usual wave equation for a free particle in the absence of any field. our theory reverts to the usual one for the motion of an electron in an electromagnetic field. The components of the 6vector curl K appearing in (4) are.Quantised Singularities in Electromagnetic Field. K0. the same K'S) but different tj^'s. It follows that if ij. The factor t]^ does not enter at all into this change in phase. written in threedimensional vectornotation. 0 . (6) Thus. since only curl K is fixed and we may use «:'s differing from one another by the gradient of a scalar for treating the different wave functions. K. Then ^ will satisfy the usual wave equation for a particle with charge —e moving in an electromagnetic field whose potentials are A = Ac/e . satisfies any wave equation. any linear combination of them cm<bm + c„^„ must also have this same change in phase round every closed curve. and this el? determines the change in phase round any closed curve. This gives a physical meaning to our nonintegrability of phase. F (7) . This is because <bm and ijj„ will both be expressible in the form (3) with the same factor elS {i. stated more explicitly. CXXXIII. but we are not obliged to do so.. equal to the components of the electric and magnetic fields E and H. It now becomes clear that the nonintegrability of phase is quite consistent with the principle of superposition. We may use the same factor et? in (3) for dealing with all the wave functions of the system. involving the momentum and energy operators p and W. VOL. that if we take two wave functions <\>m and i^n both having the same change in phase round any closed curve. 4>i will satisfy the corresponding wave equation in which p and W have been replaced by p + JIK and W — AK0 respectively. A 0 = — hje . = fE. z and t derivatives. curl« = i l H . g rad. We see that we must have the wave function <\> always satisfying the same wave equation. They are.e. (5) with similar relations for the y. whether there is a field or not.—A. From (3) we obtain iJi~J = ei^ih£ + hKx)'\>1. or. and the whole effect of the field when there is one is in making the phase nonintegrable. apart from numerical coefficients. so that the linear combination will be expressible in this form with the same e^ again. 65 where ds (a 4vector) is an element of arc of the closed curve and d S (a 6vector) is an element of a twodimensional surface whose boundary is the closed curve.
We have seen in the preceding section how the nonintegrable derivatives K of the phase of the wave function receive a natural interpretation in terms of the potentials of the electromagnetic field.* It is also contained in the work of Iwanenko and Fock. Thus this change cannot now be different by multiples of 2TZ for different wave functions. There is. Evidently these conditions must now be relaxed. Fock. being essentially just Weyl's Principle of Gauge Invariance in its modern form. . M. Nodal Singularities. The change in phase round a closed curve may be different for different wave functions by arbitrary multiples of 2TC and is thus not sufficiently definite to be interpreted immediately in terms of the electromagnetic field. namely. Iwanenko and V. 261 (1929). The present treatment is given in order to emphasise that nonintegrable phases are perfectly compatible with all the general principles of quantum mechanics and do not in any way restrict their physical interpretation.. V. The condition for an unambiguous physical interpretation of the theory was that the change in phase round a closed curve should be the same for all wave functions. R. p.' vol. The connection between nonintegrability of phase and the electromagnetic field given in this section is not new. Fock. that a phase is always undetermined to the extent of an arbitrary integral multiple of 2TC. Weyl. The more general kind of nonintegrability considered by these authors does not seem to have any physical application.' vol. Now the wave equation requires the wave function to be continuous (except in very special circumstances which can be disregarded here) and hence the change in phase round a small closed curve must be small. ' Z. as the result of which our theory becomes mathematically equivalent to the usual one for the motion of an electron in an electromagnetic field and gives us nothing new.' vol. This change was then interpreted.66 P. let us consider first a very small closed curve. p. f D. however. p. ' Z. 188. one further fact which must now be taken into account. To examine this question. H describing the electromagnetic field. It must have one definite value and may therefore be interpreted without * H. This requires a reconsideration of the connection between the K'S and the potentials and leads to a new physical phenomenon. as equal to (apart from numerical factors) the total flux through the closed curve of the 6vector E. Physik. § 3. ' C. A. 1470 (1929).f who consider a more general kind of nonintegrability based on a general theory of parallel displacement of halfvectors. 57. Dirac. by equations (4) and (7). Physik. 330 (1929). 56.
F 2 . which in turn may be associated with a direction along the nodal line. [ (H. since then its phase does not have a meaning. The total change in phase round the large closed curve will equal the sum of all the changes round the small closed curves and will therefore be 2itSw + e/fe. 2TU2W * We are here considering. As the wave function is complex.Quantised Singularities in Electromagnetic Field. H through the closed curve. This expression consists of two parts. This integer will be a characteristic of the nodal line. only magnetic flux will come into play and hence we obtain for the change in phase round the small closed curve 2im + e/hc . (8) the integration being taken over the surface and the summation over all nodal lines that pass through it. There is an exceptional case. which flux must also be small. All we shall be able to say is that the change in phase will be close to 2un where n is some integer. Its sign will be associated with a direction encircling the nodal line. j"(H. dS). that the wave function is in three dimensions. The nodal lines then become twodimensional nodal surfaces. 67 ambiguity in terms of the flux of the 6vector E. (H. H through the small closed curve. The difference between the change in phase round the small closed curve and the nearest 2TTW must now be the same as the change in phase round the closed curve for a wave function with no nodal line through it. For a closed curve in threedimensional space. the proper sign being given to each term in the sum. which can be encircled by curves in the same way as lines are in three dimensions. considerations of continuity will no longer enable us to infer that the change in phase round the small closed curve must be small. dS) which which must be the same for all wave functions arid a part may be different for different wave functions. If we now take a wave function having a nodal line passing through our small closed curve. positive or negative. its vanishing will require two conditions. so that in general the points at which it vanishes will lie along a line. for simplicity in explanation. a part e/hc. It is therefore this difference that must be interpreted in terms of the flux of the 6vector E. We can now treat a large closed curve by dividing it up into a network of small closed curves lying in a surface whose boundary is the large closed curve.* We call such a line a nodal line. occurring when the wave function vanishes. dS). The passage to four dimensions makes no essential change in the theory. however.
but the strength of such poles must be quantised. § 4. summed for all nodal lines crossing a closed surface. the quantum (x0 being connected with the electronic charge e by hcje^ = 2. The theory also requires a quantisation of electric charge. or the sum of the characteristics of all nodal lines ending there when there is more than one. Dirac. (9) This equation is to be compared with (1). If Ere does not vanish. since any charged particle moving in the field of a pole of strength u. Thus at the end point there will be a magnetic pole of strength ji. having nodal lines ending on magnetic poles. Expression (8) applied to any surface is equal to the change in phase round the boundary of the surface. are quite proper and amenable to analytic treatment by methods parallel to the usual ones of quantum mechanics. in order that wave functions describing the motion may exist. some nodal lines must have end points inside the closed surface. The value of En for the closed surface will thus equal the sum of the values of n for all nodal lines having end points inside the surface.0 must have for its charge some integral multiple (positive or negative) of e. since a nodal line without such end point must cross the surface twice (at least) and will contribute equal and opposite amounts to EM at the two points of crossing. Electron in Field of OneQuantum Pole. M. Hence expression (8) applied to a closed surface must vanish.68 P. It will perhaps help the reader to realise this if a simple example is discussed more explicitly. The wave functions discussed in the preceding section. it follows that the end points of nodal lines must be the same for all wave functions. must be the same for all wave functions and must equal — e/2n:hc times the total magnetic flux crossing the surface. Our theory thus allows isolated magnetic poles. This sum must be the same for all wave functions. = \nhc\e. These end points are then points of singularity in the electromagnetic field. The total flux of magnetic field crossing a small closed surface surrounding one of these points is 47r [i = 2rmhc/e. Since this result applies to any closed surface. I t follows that Sw. where n is the characteristic of the nodal line that ends there. A. Let us consider the motion of an electron in the magnetic field of a one .
W (11) .h*/Zm .Quantised Singularities in Electromagnetic Field. ds) round a small curve encircling this line Written is 27t. K$ = l/2r . nonrelativistically. <f> with the magnetic pole as origin. namely. except on the singular lines). We may refer all our wave functions to this set of KS. but this line may be chosen arbitrarily. KS. curl K is radial and of magnitude l/2r 2 . Let us consider a stationary state of the electron with energy W. would mean different K'S for different wave functions (the difference between any two being. tan \% (10) where KT. KT = K9 = 0. There must be some singular line radiating out from the pole along which these equations are not satisfied. 8. of course. from equations (7). If we apply the rule expressed by equation (5). we get as the wave equation for <\it . Every wave function must now have a nodal line radiating out from the origin. however. V) + i (V. This choice. We may express all our wave functions in the form (3) with the same eip. which would result in ^ being continuous.<r's will correspond to ^ ' s having a certain kind of discontinuity on this singular line. where (3 is some nonintegrable phase having derivatives K that are connected with the known electromagnetic field by equations (6). and then those wave functions whose nodal lines do not coincide with the singular line for the . We take polar coordinates r. V24* = Wiji. We express our wave function ty in the form (3). It will not. the fourdimensional gradient of a scalar. The magnetic field H. where K$ becomes infinite in such a way that (K. however. which by (9) equals \hcjer2. a discontinuity just cancelling with the discontinuity in e'3 here to give a continuous product. Hence.K2} < = W<h. K) . the wave equation is — h2/2m .. We may choose it to be the same as the nodal line for the wave function under consideration. It may now easily be verified that a solution of the whole of equations (7) is K0 = 0. K^ are the components of K referred to the polar coordinates. lies along the radial direction and is of magnitude (x0/r2. This solution is valid at all points except along the line 0 = TT. This would perhaps be inconvenient and is not really necessary. be possible to obtain K'S satisfying these equations all round the magnetic pole. 69 quantum pole when there is no electric field present. {V* + » (K.
M. The general solution of this wave equation has been worked out by I. while S6 is discontinuous for 0 = n. Tamm. From equation (12) it is evident that there can be no stable states for which the electron is bound to the magnetic pole. (13) where X is a number.70 P. that for S6 is 0 = 0. It should be observed that S a is continuous everywhere. V) = (V. Physik.472 tan2 * 6 } ^ = W^ We now suppose ^ 1 to be of the form of a function / of r only multiplied by a function S of 8 and <f> only. corresponding to which there are two independent wave functions S a = cos £0. as may easily be verified by direct substitution. ^ =/(r) S (6$.L . The lowest eigenvalue of (13) is X =  . This result is what one would expect from analogy with the classical theory. This is just what is necessary in order that both Ba and S6.. Equation (13) determines the dependence of the wave function on angle. The two ij/s that we get in this way are both on the same footing and the difference in behaviour of S a and S6 is due to our having chosen K'S with a singularity at 6 = TI. (*." . because the operator on the lefthand side contains no constant with the dimensions of a length. K) = K. A + ^ 2U ITS2 + * ' s e ° 2z i e d<f> * 4i sin 6 S6 30 sin 8 3<£ ' ~ i t a n 2 i 6 } S = . The general eigenvalue of (13) is X = w2 + 2m + \.*S.* * Appearing probably in ' Z. i.t a n 2 A0.£.L {V2 + isec2 *e £ .e. A. It may be considered as a generalisation of the ordinary equation for spherical harmonies. The nodal line for S 0 is 0 — 7r. 4rso that equation (11) becomes . Sb = sin 0 e'*. = zsz sec2 16 ± r sin 0 09 4r d<£ The values (10) for the K'S give K" = /o. . its phase changing by 2n when one goes round a small curve encircling the line 0 = n. 2 = —. when multiplied by the e'p factor. may give continuous wave functions i . Dirac. This requires a sin e .
Quantum mechanics. but is merely a generalisation of the possibilities of representation of these abstract symbols by wave functions and matrices. Conclusion. § 5. when developed naturally without the imposition of arbitrary restrictions. there appears to be no possibility of modifying the theory. The theory leads to a connection. such as (1). so presumably the explanation of (1) will require some entirely new idea. and this is possible only when there are no isolated magnetic poles. This new development requires no cJuinge whatever in the formalism when expressed in terms of abstract symbols denoting states and observables. instead of a purely electronic quantum condition. leads inevitably to wave equations whose only physical interpretation is the motion of an electron in the field of a single pole. The theory would now run quite parallel and would lead to the same condition (9) connecting the smallest pole with the smallest charge. we have to introduce the electromagnetic potentials. The object of the present paper is to show that quantum mechanics does not reallv preclude the existence of isolated magnetic poles. the present formalism of quantum mechanics. The experimental result (1) shows that there must be some . There remains to be discussed the question of why isolated magnetic poles •ire not observed. This would require the introduction of the electromagnetic potentials B satisfying E = curl B. between the quantum of magnetic pole and the electronic charge. 71 Elementary classical theory allows us to formulate equations of motion for an electron in the field produced by an arbitrary distribution of electric charges and magnetic poles. H = . c ot to be used instead of the A's in equations (6). as it contains no arbitrary features. as we did in § 4. Under these circumstances one would be surprised if Nature had made no use of it.grad B 0 . is derived from the Hamiltonian form of the classical theory and therefore is applicable only when there are no isolated magnetic poles. we could equally well consider the motion of a pole in t he field of fixed charge. equation (9). However.——\. On the contrary. 1 nstead of discussing the motion of an electron in the field of a fixed magnetic pole. If we wish to put the equations of motion in the Hamiltonian form. as it is usually established.Quantised Singularities m Electromagnetic Field. however. The theoretical reciprocity between electricity and magnetism is perfect. It is rather disappointing to find this reciprocity between electricity and magnetism. namely.
This means that the attractive force between two onequantum poles of opposite sign is (137/2)2 = 4692^ times that between electron and proton. not \x0 — e. .290 72 F. This very large force may perhaps account for why poles of opposite sign have never yet been separated. cause of dissimilarity between electricity and magnetism (possibly connected with the cause of dissimilarity between electrons and protons) as the result of which we have. Hitciien. but [x0 = 137/2 . Twyman and C. e. S.
x. all physical fields decrease exponentially outside this kernel. In such a nonAbelian theory one can only imagine nonoriented strings. SO(3). 1.0)l(x2+y2). then A(x)cc(y.Nuclear Physics B79 (1974) 276284. Geneva Received 31 May 1974 Abstract: It is shown that in all those gauge theories in which the electromagnetic group U ( l ) is taken to be a subgroup of a larger group with a compact covering group. if they have the same direction they may only join to form an even tighter string. Their mass is calculable and of order 137 My. in a superconductor. 't HOOFT CERN. Here y is the angle about the z axis: ft(0)=n(27r)= 1 . Outside the kernel we do have a transverse vector potential/1. Or: magnetic 7 monopoles do not occur in the system. let us suppose that the electromagnetism in the superconductor is in fact described by a unified gauge theory. but there it is rotationfree: if we put the kernel along the z axis. like SU(2) or SU(3). (1. Now. Also it is easy to see that these strings are oriented: two strings with opposite direction can annihilate. It is obvious that such a string cannot break since we cannot have an end point: it is impossible to replace a rotation over 2 T continuously by £2(<p) * 1. [1]. behave very much like the Nambu string [2]. NorthHolland Publishing Company MAGNETIC MONOPOLES IN UNIFIED GAUGE THEORIES G. in which the electromagnetic group U(l) is a subgroup of. the .1) A(x) can be obtained by means of a gauge transformation £2(i/>) from the vacuum. because a rotation over 4n can be continuously shifted towards a fixed H. Introduction The present investigation is inspired by the work of Nielsen et al. say. who found that quantized magnetic flux lines. Their solution consists of a kernel in the form of a thin tube which contains most of the flux lines and the energy. genuine magnetic monopoles can be created as regular solutions of the field equations. where My is a typical vector boson mass..e. What happened with our original strings? The answer is simple: in an SO(3) gauge theory magnetic monopoles with twice the flux quantum (i.
In a nonAbelian theory with compact covering group. It can be obtained from the vacuum by applying a gauge transformation A: ^=VA. oriented in the same direction. occur. 1. until it shrinks at the bottom of the sphere. with a magnetic flux <> entering at one spot J (see fig. From now on we shall dispose of the original superconductor with its quantized flux lines.4. on the contour C0 in fig. (1. which transform according to i// > ii eniA . Immediately around that spot. even rotations over lit Fig.2) This A is multivalued. Now we require that all fields. t Hooft. (1. That these monopoles are possible. so <> must be an integer times 2ir: we then have a comJ plete gauge rotation along the contour in fig. Two of the original strings. a rotation over 4n may be shifted towards a constant. In other theories. as regular solutions of the field equations. 1. can be understood in the following way.G. we must have a magnetic potential field . with §{A • dx) = 4>. 1). 1. however. We consider free monopoles in the physical vacuum.. for instance the group 0(3).3) to remain single valued. We deplace it from Co to Ci. In an Abelian gauge theory we must necessarily have some other spot on the sphere where the flux lines come out.4] value). . Magnetic monopoles 277 Schwinger [3. because the rotation over 2kn cannot continuously change into a constant while we lower the contour C over the sphere. etc. We require that there be no singularity at that point. without singularity: we may have a vacuum all around the sphere. can now annihilate by formation of a monopole pair [5]. Imagine a sphere. The contourC on the sphere around the monopole.
For such a rotation one can. 2. 1 will be spread over the whole sphere. As we shall see in sect. this Q. There are two classes of possibilities. say. At one side of the sphere (cos 6 >• 1) we have a rotation over 4n. however. where M w is the * As we shall see this vacuum will still contain a radial magnetic field.^) = cos0 +sin^( /° ' \ 1. in isospin space.278 G. One easily checks that ftftt = 1. then still monopoleantimonopole pairs. Now.4) Now consider one rotation of the angle y over 2n. nor is there the need for a Dirac string. This is why a magnetic monopole with twice or sometimes once the flux quantum is allowed in a nonAbelian theory. are legitimate solutions of the field quations. If we want to be conservative and only permit the normal boundary condition at infinity. This is how we were led to consider solutions of the following type to the classical field equations in a nonAbelian Higgs—Kibble system: a small kernel occurs in the origin of three dimensional space. 2. unfortunately. this £7 rotates over 4ir (the spinor rotates over 2n). At 6 = 0.5) In the usual gauge theories one normally chooses the gauge in which the Higgs field is a vector in a fixed direction. this leads to a new boundary condition at infinity. (1. to which corresponds a nontrivial solution of the field equations: a stable particle is sitting at the origin. if the electromagnetic group U ( l ) is a subgroup of a gauge group with compact covering group. along the positive z axis. tp). we take as a gauge condition that the Higgs field is £1(6. with Higgs fields pointing in the z direction. for instance. is a constant. Outside that kernel a nonvanishing vector potential exists (and other nonphysical fields) which can be obtained from the vacuum* by means of a gauge transformation £1(6. excludes the popular SU(2) X U ( l ) model of Weinberg and Salam [6]. Magnetic monopoles may be shifted towards a constant. At 6 = v. we can construct monopoles with a mass of the order of 137 M w . t Hooft. That. based on SO(3). It will be shown to be a magnetic monopole. This is because the incoming field in fig. . The model We must have a model with a compact covering group. which goes to unity at the other side of the sphere (cos0 »• —1). (i) In models of the type described by Georgi and Glashow [7]. take the following SU(2) matrix: /e'V ° \ n(e.ip) times this vector. (1. arbitrarily far apart. There is no singularity anywhere in the sphere.
In the Georgi—Glashow model. apart from the obvious angle dependence.9]: «2a>2sF2> v2 = \\F2. We choose the first possibility for our sample calculations. M w < 53 GeV/c2. (2. it implies that the Higgs field must have at least one zero. (2.0. r2a=r2. sin d sin ip. (2. Then the monopole mass would be 137 times the mass of one of the superheavy vector bosons. As one can easily verify.1) DiiQa = Z»Qa *eabcWtQc (22) W° and Qa are a triplet of vector fields and scalar fields. + . We now ask for a solution of the field equations that is timeindependent and spherically symmetric. cos 0).1) we get F(sin0 cosip. Magnetic monopoles 279 mass of the familiar intermediate vector boson. (2.5) to the isospinone vector F(0. We choose the parameter n2 to be negative so that the field Q gets a nonzero vacuum expectation value [6. respectively. (1.z). t Hooft. If we apply the transformation £1 of eq. The Higgs particle has a mass: MH=VXF. \i?Qa ^(Q2a)2 .7. Introducing the vector ra=(pc. We take as our Lagrangian: £ = iG%G% where G% = ^K .6) We shall take this isovector as our boundary condition for the Higgs field at spacelike infinity.5) We are interested in a solution where the Higgs field is not rotated everywhere towards the positive z direction.G. (2. Weinberg [8] proposed SU(3) X SU(3) which would then be compact.y.3) Two components of the vector field will acquire a mass: MwV2=eF. because it is the simplest one.{D^Dfi.3„W/« + e eabcW*Wcv . (ii) The WeinbergSalam model can still be a good phenomenological description of processes with energies around hundreds of GeV. This zero we take as the origin of our coordinate system.but may need extension to a larger gauge group at stilLhigher energies.4) whereas the third component describes the surviving Abelian electromagnetic interactions.7) . (2.
(2. as some negative power of r: W(r)^ar".t) = raQ(r).9) where the constant has been added to give the vacuum a vanishing action integral.6W2 2eW + XF2r222Xr424XF4 (2.11) H^f+H = r2[4r^ + 12W +6er2W2 +2e2r4WJ +2er2Q2 +2e2rAWQ2} . Magnetic monopoles we can write Qa(x. (2. In terms of these variables the Lagrangian becomes L = Jj2d*x = 4n j r2 dr\r2 0 L (*£¥ V ^/ 4rW ^ . far from the origin. let us concentrate on the boundary condition at r * °°. + 2e2a^r6~3" (2. the fields are .9) we find the Lagrange equation (2. W > . substituting (2. r ) = e^abrbW(r). (2. From the preceding arguments we already know that we must insist on QOOF/r. 3.10) since our system is stationary.14) + 2 e F 2 r 2 + 2e2aF2r4" The only solution is n = 2. The field W must behave smoothly.12) (2. 2. (3 n)(42n)ar2n r y4nar2~n + \2ar2~n + 6ea2r'i~2n .8) where e^ab is the usual e symbol if 11 = 1. 't Hooft.280 G.12).13) So. From (2. and eAab = 0.15) So. (2. The energy of the system is then given by E = L.11) and (2. a =l/e. The field equations are obtained by requiring L to be stationary under small variations of the functions W(r) and Q(r). Before calculating this energy.
except where Qa = 0. which will yield the usual definition in the gauge where the Higgs field lies along the z direction everywhere. we get (see the definitions (2.— W * • 3 (22°) Again. Q a (*.) From (2. (2.0. our solution is a magnetic monopole. the e symbol has been defined to be zero as soon as one of its indices has the value 4. In sect. Qa = 121(0. in particular the electromagnetic ones. if after a gauge rotation.22) n integer. is satisfied.17) because. ^ . Q A . . this is one other way of understanding the possibility of monopoles in this theory. (2. f) . t) + .^ ^bcQa(D.Qb)(DvQc) . =. FM„. It satisfies Schwinger's condition eg=l (in units where h = 1). So. then we have there as one can easily check. (2. we must first give a gauge invariant definition. as we expected. (Observe that the definition (2.e^abrb/er2 . Magnetic monopoles 281 Wftx.18) (2.16). (2.16) Now most of these fields are not physical. we show that in certain cases only Dirac's condition eg = \n ..Frjr . t Hooft.1) everywhere within some region.21) Hence.. Hence (2. there is a radial magnetic field Ba=rJerK with a total flux 4n/e .19) F »v = . We propose: *•„* = £.G.17) satisfies the usual Maxwell equations. To find the physically observable fields.2)): 2 ( C . D„Qa = ^)iQa+eeabcW^Qc=0. 4. however.
X4W4 + 1.1 for j3 = 0. as a consequence of our approximation procedure. On the other hand.*V + /3 (3.10) we find that the energy E of the system is the minimal value of 4wAfw e 2 / 0 x2 dx [*2 ( < g ) 2 + 4XW g + 6W2 + 2 *2 W 3 + . In other models it will in general be the mass of that boson which corresponds to the gauge transformations of the compact covering group: some of the superheavies in Weinberg's SU(3) X SU(3) for instance. * These values may be slightly too high.\$x2q2 + /3. and that their properties are predictable and calculable. Conclusions The relation between charge quantization and the possible existence of magnetic monopoles has been speculated on for a long time [10] and it has been observed that the gauge theories with compact gauge groups provide for the necessary charge quantization {11 ]. (3.9) and (2. Magnetic monopoles 3. The mass of the monopole Let us introduce dimensionless parameters: w = W/F2e .+ \q2 + 7x2wq2 +x4w2q2 .3) where C(0) is nearly independent of the parameter (3.1 to 1. q = Q/F2e .282 G. 3=X/e2=M^/M^. x=eFr. Nevertheless. Only in the Georgi—Glashow model (for which we did this calculation) is the parameter M w in eq. solutions of the'field equations with abnormally rotated boundary conditions for the Higgs fields have also been considered before [1. it had escaped to our notion until now that magnetic monopoles occur among the solutions in those theories.2 f*JLJ + xq p. 4.12].1) From (2. We found that the mass of the monopole (which is equal to the energy E since the monopole is at rest) is Mm=^MwC(0). t Hooft. (3. .44 for 0= 10*. It varies from 1. (3.3) really the mass of the conventional intermediate vector boson.2) The quantity between the brackets is dimensionless and the extremum can be found by inserting trial functions and adjusting their parameters.
Evidently. [4] J. D5 (1972) 1962. Dirac. New York. Parisi. Zumino and D. Phys. Nuovo Cimento 69A (1970) 457. Phys. and the predicted monopole mass will be again much higher. Phys. Olesen. Niels Bohr Institute preprint. and B. Schwinger. Rev. Rev. Schwinger's arguments do not hold for this theory [13]. Zumino. Magnetic monopoks 283 Our way of formulating the theory of magnetic monopoles avoids the introduction of Dirac's string [3]. Nielsen and P. Rev. 1970) p. but this is certainly not a general phenomenon.M. Proc.A. If Weinberg's SU(2) X U(l) model wins the race for the presently observed weak interactions. 74 (1948) 817. one may introduce isospin \ representations of the group SU(2) describing particles with charges ±\e. Soc. where Aq is the chargedifference between members of a multiplet. Gross for interesting discussions.B. [2] Y. on symmetries and quark models. Detroit. 144 (1966) 1087. In Weinberg's SU(3) X SU(3) the monopole quantum is the Dirac one and in models where the leptons form an SU(3) X SU(3) octet [14] the monopole quantum is three times the Dirac value (note the possibility of fractionally charged quarks in that case). J. Int. We expect no fundamental problems in calculating quantum corrections to the solution although they might be complicated to carry out.L. 1779(1973). Phys. Georgi and S. Glashow. but only Dirac's condition Qg = i > where q is the charge quantum and g the magnetic pole quantum. Lectures given at the 1973 Nato Summer Institute in Capri. then we shall have to wait for its extension to a compact gauge model. Gervais and B. Proc. Weinberg. t Hooft. We do have.L. Sakita. in our model &qg= 1 . The prediction is the most striking for the Georgi—Glashow model. in spite of the fact that we have a completely quantized theory. Letters 30 (1973) 716. Nambu. [6] S. [3] P. Copenhagen (May 1973). Finally. We thank H. [8] S. 1969 (Gordon and Breach. L. one important observation.G. Letters 19 (1967) 1264. In that case our monopoles do not obey Schwinger's condition. 269. although even in that model the mass is so high that that might explain the negative experimental evidence so far. Susskind. Phys. In the Georgi—Glashow model. Rev. [7] M. Phys. [5] G. Columbia University preprint CO227129. Rev. CERN preprint TH. Conf. Weinberg. References [1] H. Letters 28 (1972) 1494. Roy. Strubbe for help with a computer calculation of the coefficient C((5). A133 (1934) 60. Rev. . B.
P.S. Letters 13 (1964) 508. Phys. Rev. Dl (1970) 2360. 711. [13] B. Magnetic monopoles [9] F. G. . Private communication. Magnetic monopoles: an updated bibliography. [14] A.N. and State University preprint VPIEPP735 (October 1973). University of Maryland preprint (November 1972).C. [11] C. [12] A. Rev. Letters 13 (1964) 585. Phys. Hagen and T. Zichichi (Acad. Phys.284 G. Stevens. Rev. ed. Inst. New York and London) p. Higgs. Kibble.W. Phys. Rev. Letters 12 (1964) 132. Virginia Poly. 145(1966) 1156. Englertand R. 1966 Int. Brout. Press.R. School of Physics. Neveu and R.M. Rev. A. C. Erice. Strong and weak interactions.W. Salam and J. Phys. Phys. t Hooft. Dashen. [10] D.B. Letters 13 (1964) 321. Yang. Pati. Guralnick. Zumino.
The solution of the classical equations is (6) «. the model under consideration has one m a s s l e s s vecton and two massive vectons.x u(r)r~l . which mean tances. owing to the inhomogeneity of the distribution of that the vacuum is perturbed only within a finite volume. 3. we obtain the mass spectrum and to add the YangMills Hamiltonian to (5). the inhomogeneity of the directions then becomes physically unrealizable and makes no contribution to the energy. . r yj2 — . Over the vacuum there is a singleparticle state with mass V"2(i. D. The first is to connect to the hedgehog a YangMills field. As is well known. By virtue of the gauge symmetry of the second kind. the results consists in the fact that each stationary regular solution of the classical equations of motion corresponds in quantum field theory with weak coupling to its own set of ext r e m o n s . a = l. the second is the zeropoint oscillation energy. in this model there a r e three types of particles with anomalously large m a s s e s .4 c 2 ^ 2 <j>a = o (KV (5) The equation for the extremal v V . i . Thus. = 0. 6.' /.300 Particle spectrum in quantum field theory A. = . No. 1974) ZhETF Pis.(x) In this model.. 1974) We wish to call attention in this article to a hitherto uninvestigated part of the spectrum of ordinary Hamiltonians used in quantum field theory.A .n} (4) p + — \n(i 2VT .* S 4>\ has a solution d. e .. We shall show that extremons exist in t h r e e dimensional models. Polyakov L. 2. the first term is the potential energy at the equilibrium point.V u • « ' . 1. 194 JETP Lett.* + ( ( . . (1) * (*) .(x)4c(z) + d>(x) Vu^a . inasmuch as the isovector at a given point of space is directed along the 4>" + ^ <j>c \$l . No. with Hamiltonian H = — t dx 2 . 6. We call this the "hedgehog" solution. to make the substiU U X (3) tution To calculate the spectrum of the vibrational energy levels near the considered equilibrium point. Bd. There is another extremal of the potential energy in (1). We consider the theory of the Higgs isovector field 0„(x).. the masses of which can in principle be calculated. the vacuum is filled with the Bose condensate <f>2 = p.J U 3 =0 The mass of the resultant extremon is of the order of " ^ ~T m~v 2 2 v/s (where m ir =g u (°°) is the mass of the vector bosons). — 2V2^3 y/S+2 t. 2 . 20. As the first example we consider the model of a selfinteracting scalar field in twodimensional spacetime. " • 2 . Diagonalizing the obtained quadratic Hamiltonian. The solitary hedgehog is not an extremon.» H = l 2 " 1 2 f v^*" j' 2 . 0 (2) radius vector.. and the third is the excitation energy. However. d e termined from the equation where u is subject to the equation . 1974 where u and a satisfy the equations + _ „ ' + {^2B2a2(r)) U . Red. Vol.^ . In formula (4). ufr ) = c o n s t r . Landau Institute of Theoretical Physics (Submitted July 5. constant 0 is not the only stable equilibrium state. the directions of the field <pa." In the generalized formalization. 430433 (September 20. ! < and neglect the t e r m s 0 3 and $ 4 in the Hamiltonian (the latter is valid if \ « ju2). September 20. which represents small oscillations of the constant condensate. M. since its energy diverges linearly at large diswith boundary conditions 02(±«>) = p?/\. which therefore turns out to be finite. We call these objects "extremons. .2/X. If the Copyright © 1975 American Institute of Physics 194 . 20. + n ! 0 a . There a r e two ways of The solution of (2) is overcoming this difficulty. ^ O " f °° ) = ji / > A '. ' . we write d.
There is no doubt.B.B. 45(1973). and L. LeeandG. " It is easy to construct hedgehogs in which all the components of the gauge field a r e massive and concentrated within the region l/mr. then the hedgehog has by virtue of (6) a magnetic c h a r g e . (8) ^This circumstance was pointed out to me by L. B61.. IPM (Inst. In particular. where R is the distance between the pair components. To this end it suffices to consider an isotensor or isospinor Higgs field. D .) Preprint. L . B . Math.301 first of them is identified with the photon. After completing the work. Wick. D. B. Nucl. L . that a formula of the type (8) for the energy should lead to growing Regge trajectories. n . in press. Therefore the validity of (8) depends on the hypothesis concerning the character of the hedgehog rotation. I am grateful to V . 2 A rigorous justification of (7) calls for the solution of the equation for the extremal with allowance for the fixed angular momentum. 1974.Yu. Lett. Okun». 3 T. To stabilize this state it is necessary to consider levels with zero angular momentum L (7) The mass spectrum is given by H*(L) = const L . Another possibility for the construction of an e x t r e mon is formation of a hedgehogantihedgehog p a i r . Columbia University Preprint. COO227120 (1974).C. Okun*. Nielsen and P. then we obtain in place of (8) the formula M*(L) = const i . Faddeev. Ideas very close to those described above were developed in 11 " 4 ' but it seems to me that both the analysis method and the results obtained above a r e new to some degree. Kobzarev. and L . We have derived such an equation. It is easy to show that the pair energy is E=AR. 'H. Phys. and in the latter case it has a more complicated form and will be described elsewhere. In the former case the solution should be sought in the form if it is assumed that the region of space contained between the hedgehogs takes part in the rotation. Okun'. I obtained a preprint by t'Hooft. Appl. Thus. which contains a similar result. Zel'dovich. Ya. in the interpretation of the elementaryparticle spectrum on the basis of field theory one should bear in mind the many new possibilities that a r e afforded by taking the extremon states into account. Olesen. . I. but a r e unable to solve it exactly. however. Faddeev for useful discussions. D. Phys. 4 L. Berezinskii.
It is the purpose of the present article to point out that it is easy to build up classical field theories allowing for vortex lines (or similar stringlike structures) with the property of having equations of motion identical with those of the Nambu dual string [2]. On the other hand it is not unreasonable to expect that some field theory describes relativistic physics (presumably including strong interactions and certainly electromagnetic interactions). It turns out that this strong coupling limit is "super quantum mechanical" in the sense that the typical action of the theory is very small in comparison with Planck's constant. There are several motivations for being interested in field theories for dual strings: (i) We have good reasons to believe that both field theory (of a kind which is so far not known) and dual strings (with some yet unknown degrees of freedom) are in fact realized in nature. University of Copenhagen. For example. in analogy with the vortex lines in a type II superconductor. It is therefore likely that nature has decided to merge some field theory with some dual string structure. These vortex lines can approximately be identified with the Nambu string. crossing symmetry which is crucial for the Veneziano model.6 1 . Introduction and motivation The many good experimental and theoretical features of the Veneziano model seem to suggest that an underlying string structure [1 —4] of hadronic matter is a likely possibility. (ii) One may hope that by building some field theory for the dual string one is .B. we show that a Higgs type of Lagrangian allows for vortexline solutions. 1. The spectrum of the Veneziano model would then be brought into contact with a local field theory. It is therefore of interest to see if one can cook up a local field theory which gives string structures behaving like dual strings.Nuclear Physics B61 (1973) 4 5 . DK2100 Copenhagen 0. is based on ideas taken from field theory. OLESEN The Niels Bohr Institute. Denmark Received 6 June 1973 Abstract: We call attention to the possibility of constructing field theories for the dual string. NIELSEN and P. In the strong coupling limit we speculate that the vortex lines make up all low energy phenomena. The width can be computed in terms of the parameters of the specific field theory model that one uses. As an example. The string description (and thus also the dual string equations of motion) are obtained only in the approximation where the radius of curvatures of the string is much larger than the width of the string. NorthHolland Publishing Company VORTEXLINE MODELS FOR DUAL STRINGS H.
The Higgs model may also be considered as a relativistic generalization of the GinzburgLandau phenomenological field theory of superconductivity (see ref. In sect. Dual strings able to secure that the good properties of the former are inherited by the latter. and it is exactly this fact which allows us to connect the Higgstype of Lagrangian with the dual string. for example one might hope that by choosing a field theory with positive definite Hamiltonian one might avoid tachyons in the corresponding dual string model. P. the question is as to whether the unphysical amount of dimensions needed in dual models [4] may be related to internal symmetries of the hadrons. It is . This model has been used by Higgs to illustrate the Higgs mechanism. Thus. at least not if this is done in the sloppy way of this paper. 3 we show by a very general argument how the vortex solution leads to a Nambu Lagrangian [1]. d = 10 in the NeveuSchwarz model) into a field theoretic language. it is still interesting that in order to understand hadronic structure in the sense of dual strings one has (in some sense) to consider the quantum of action h to be very large. In sect. the possibility of understanding what happens when a couple of strings collide seems to have a chance in field theory. 2 we consider the Higgs Lagrangian in the static case. Nielsen. the typical action in the theory is much smaller than the fundamental quantum of action given by Planck's constant h. Although this may be bad from the point of view of quantization. In the GinzburgLandau case one knows the existence of a vortexline solution. 4 we deal with the problem of getting the width of the vortex line sufficiently small in order that we can get a sufficiently good approximation to the dual string. Also. Olesen. whereas field theory allows many possibilities. The relevance to dual models of an Abelian gauge field was first pointed out by us in ref. That is to say. This argument is very independent of the specific features of the Lagrangian. (iv) We may use field theory to get ideas for how to modify the existing dual (string) models to get perhaps some day the right model. (iii) One may hope to understand the features of dual strong models better by considering the corresponding field theories. 4) that the limit in which the Higgs Lagrangian gives the dual string solution is a sort of superquantum mechanical limit.B. We recapitulate the most relevant features of this theory. and it would be of great interest to translate the requirement of a critical dimension (d = 26 in the conventional model. It turns out (see end of sect. In particular. [6]). In sect. this hope is apparently not supported when quantum mechanics is taken into account. [5]. But we can still hope that a more careful treatment of quantization can resolve this difficulty. and we discuss how the vortex solution comes out. As the main example of a field theory with a dual string structure we consider the theory of an Abelian gauge field coupled to a charged scalar field.46 II. provided we identify the vortexline with the dual string. and it turns out that this Lagrangian becomes identical to the GinzburgLandau free energy in the theory of type II superconductors. The trouble with generalizing dual models is that they are so tight. As we shall see by an example.
.2c 2 0 + 4c 4 0 2 0* .C 4 l 0 l 4 . P. (2.1) We are now looking for a solution of the vortex type*. The main new result is the identification of the GinzburgLandau theory with the static solution of the Higgs type of Lagrangian (2. ref.=/M=iM**aM0*3M0*)+«2/lM0**. An example of a static vortex solution of the GinzburgLandau type In this section we shall consider a special example of a field theory with a vortex solution. 2. To see that this is the case. [5]. 5 we give a very brief and preliminary discussion of the problem of quantization. 3"^M1. Dual strings 47 shown that we have to consider the strong coupling limit.B. We start from the Lagrangian •G = . Writing (2. we use that the flux is given by * =/ F ^ d ^ = ^ where da M W d / . (2. In sect.4) over a closed loop without any current.5) 0=l0le'' x 'V' we get from eq. (2. and it is thus necessary that the flux is quantized.3) 1 A* 1 A =— .^ .4) is a twodimensional surface element in Minkowskispace.7) * Most of the results obtained in this section are known in the theory of type II superconductors (see e. and the equations of motion are (3M + ieAJ24> = .g. In an appendix we deal with the case of a YangMills Lagrangian. We want to identify the vortex line with a dual string.^ a M x W d .i ^ M ^ ' M l ' + T l O M + ^ M ) 0 l 2 + c 2 l 0 l 2 . [6]). Olesen. Then 4) = (J)/lM(jc)dxM = . In this case the field F has a simple meaning: the field F 1 2 measures the number of vortex lines (going in the 3direction) which pass a unit square in the (12)plane.H.1).rb x(26) M e2 I0I 2 e M Next let us perform the integration in (2.2) (23) (2. This interpretation is identical to the one proposed by us in ref. . Nielsen. v ' i . (2.
(2. We look for a cylindrically symmetric solution.14) If we treat 10 I as a constant.4>(/) = 1 f (r\A\). i. with c a constant of integration.10) With cylindrical symmetry around the zaxis the equations of motion (2. We write rX e A(r) = —^\A(r)\. with axis along the zdirection.11) then gives li/l = d 0 l K o ( e l 0 l r ) —• °. (2. so that \H\ = ~ f. We shall be content with a solution of the type where (2.° I 0 I — const (for large r) .. Let us consider the static case. Thus.11) (2. *o=T (2 8)  Thus. with a gauge choice A0 = 0.2) and (2. (2. Nielsen.l /  S I e . 2n/e being the quantum.(7jr<'ul>) + l *  2 ( u l e 2 ~ 7 ) .48 H.12) f. The flux is given by *(r) = 27rrL4(r)l.e.9) where ez is a unit vector along the zdirection. We still have to show that the equations of motion (2. (2. One finds.> The exact solution of these two equations has so far not been obtained analytically.e l 0 1 ' ' + lower order. The only general requirement on the phase is that 0 is single valued. (2.2) and (2. Dual strings The line integral over the gradient of the phase of 0 does not necessarily vanish. * = «*<>.3) allow a stringlike solution. \A\=—+KAe\(t>\r) er e e l&rrf 2e101r e_e 0 —*• — + r>=» ' " + lower order terms.B.3) give v2 2 c + '4 ^ u„4 I0l 2 c ^2 l0l = O. x varies by 27rw (n = integer) when we make a complete turn around a closed loop.13) can be solved without further approximations. Olesen. the flux of vortex lines is quantized.15) Eq. (2. P. (2. eq.16) . (2 13 .
H. p(x) ~ e . whereas the second derivative is 2c4l0ol2=2c2. I0I = 0 O = ] / T T • is the vacuumvalue of the field I 0 I.^ ' .20) (219) which is the mass square of the scalar particle in Higg's mechanism. Then the oscillations in the potential are 2c 2 p(x) 2 leading to a solution for p of the Yukawatype. Olesen. (2. (2. The first derivatives of the potential Kc200*+c4020*2).. We next define the characteristic length X (called the penetration length in superconductivity). We then define a new characteristic length %.17) gives the minimum of the potential.22) (2. i. (2.12) is then approximately satisfied if 101( ^ . £ measures the distance that is takes before the field 10 I reaches it vacuum value. *=^=(2 (2.23) "24) Thus. Let us write l0l = 0 o + p ( x ) .B. where p(x) give the fluctuations around the vacuum. Nielsen. (217) with c 2 and c^ being large so as to take care of deviations of IA I from l/er. (2.21) (2.e. Dualstrings 49 Eq. P. . e\<f\ K e2c2 X thus measures (see eq. To estimate the variation of 10 I we notice that eq. vanish.16)) the region over which the field H is appreciably different from zero.
It is seen that if £ and X are of the same order of magnitude. This is simply due to the fact that the Higgs Lagrangian is a relativistic generalization of the GinzburgLandau Lagrangian. 1 we have illustrated the behaviour of the fields 17/1 and I 0 I.50 H. so that it has a meaning to separate out the special . In this section we shall concentrate on the vortex contribution to the Lagrangian (3.15) is determined by the requirement that the flux $(/) = 2nr\A(r) I shall go to zero for % <r <\.^ e s n a " a s s u m e t n a t the other solutions can be effectively decoupled from the vortex solution.e I 0 I. 1.1) allows a stringlike solution. (2. Now. The Nambu Lagrangian from vortexline structure In the preceeding section we pointed out that the Higgstype of Lagrangian J2 = ^FF^+\\(d +/Mu)0l2+c. The constant of integration c introduced in eq. The main point of this section is thus that the Higgs type of Lagrangian (2. (2. 101 (3. By a suitable choice of parameters we can arrange that the width of the vortexline is much smaller than the radius of curvature of the vortexline. Dual strings Fig. J2vortex s a v . (3. P. Nielsen.1).26) 3. In the next section we shall discuss how to handle some of the other solutions.1) has solutions of the vortex type. in addition to the vortex solution. In fig. and the extension of the string is given by £ ~ X. An example of the behaviour of the fields I H\ and I 0 I for a vortex solution. then we have a welldefined vortex line. The vacuum state is described by H = 0. which is well known to have vortex solutions. Of course.1) certainly has other solutions. Olesen. and l0l = V l ^ / ^ . for 0 < e I 0 I r < 1 we have i e I 01r (2.B.25) and consequently c = . or a well defined string.
It therefore follows that i?vortex is Lorentzcontracted in the transverse direction..H t H i r I)2 • Introducing a different parametrization of the vortex.1) is a smeared out 5function. For simplicity we shall assume that the endpoints are at spatial infinity.3) w / . Dual strings 51 In the last section we saw that by choosing the characteristic length X sufficiently small.B. All we need is that a vortex solution exists. is a constant. which is only nonvanishing along the vortexline. (3.2 we have illustrated the vortexline. Goldstone. Nielsen. Let ds be the element of length along the vortexline. °l provided one chooses a frame [4] where the parameter T is identified with the time /. Of course. and vice versa.6) .dx/ds(dx/dt and hence • dx/ds) (3. X ~ if.2) due to the motion of the vortexline in the transverse direction. x(s. the field F (x) is nonvanishing only in a small region. The quantity — (\)F2 in eq. £ is itself a smeared out 6function. Infig. \<P I practically speaking vanishes when Fflv is nonvanishing. it follows that the action is given by (we ignore constants of proportionality) S„ vortex Now vl = dxfbt . Cortex .4) fd'x Cortex "fdtdsy/T^.e. £ is also relativistically invariant. whereas the field I 01 is nearly always equal to its vacuum value \/c2 / 2 c 4 . (3. Olesen. or (perhaps better) that the vortex is a closed loop. (3.1). which is nonvanishing only along the vortexline. Since the "transverse length" of the string. Taking £ ~ X. t) this leads to the Nambu action [2] (discussed in detail by Goddard. The field F^^x) therefore acts as a smeared out 6function. P. (3.H. Rebi and Thorn HI) Cortex/ l as d f / Zofas/Tv{. It is rather clear that the arguments presented above do not depend on the details of the Lagrangian (3.1) therefore also acts as a smeared out 5function. apart from the region in space where F)lv is nonvanishing. Since the vortexcontribution to the Lagrangian (3.V T ^ f . i. t) = x(a.
It must be emphasized that our considerations are entirely classical. 2. He showed that the theory by i. Thus we have obtained the following rather remarkable result: the action of the vortex is proportional to the area of the surface swept out by the vortex in space and time. only a special example) with the dual string. P. (This would be a very inappropriate gauge choice for a vortex state). The strong coupling limit Our example. This result forms the basis of our identification of the vortex solution of the Higgstype Lagrangian (which is. namely the field theory of the LandauGinzburg type. and so far no quantum effects are included. Lorentz contraction of the vortex in the transverse direction.1) . is the one used by Higgs to illustrate the Higgs mechanism.B. Choosing this gauge and putting 0(*) = 0 o + p ( x ) . of course.n appropriate gauge choice is revealed to be a theory of a massive scalar and a massive vector meson. 4. In fact he chooses a gauge where the phase of the charged scalar field vanishes.309 52 H. Nielsen. where 0Q is given by 00 = f ^ . Dual strings Lorentz contraction: X — XN 1v^ Fig. (42) (4. described by the Nambu Lagrangian. Olesen.
% = m. Goddard. r Z64 C (4. 2. Here a is the universal slope for the state of the string with two ends.9) The integral JQ KQ(y)2 y dy converges because K0 (y) behaves no worse than a logarithm for 7 > 0 and decays exponentially for/ *• °°. which for a dual string model is \fcx. Rebbi. The magnetic energydensity along the vortex string is oo oo oo energy density = \ j 0 \H\ 2nrdr 2 = \^e 2 J KQ{e^Qrf 0 2nrdr = TT4> f 0 2 K0(y)2ydy =^ \ f 4 0 KQ{y)2ydy.B. that is to say are thin*. the energy density along the dual string is 7 i 1 = —.6) (4.1 . 1).1 (4.5) (4. c2.H.8) 2na. (4.' 2na' \J\ — x> where vx is the transverse velocity of the string. % is the radius of an inner core in the string where 10 I deviates appreciably from its vacuum 0O> while X is the radius of the string determined by the width of the flux bundle (compare with fig. We can thus make a classical estimate of the connection of the universal slope a to the three parameters.4) and a scalar meson with mass m s = s/2e2 interacting with each other. c4 and e of the LandauGinzburglike Lagrangian (2. . Olesen. In the case X > £ discussed above. (4. and Thorn [4].3) This form is easily seen to describe a vector meson with mass mv = e0 = e l/ 2 tf — . Nielsen. P. Dual strings S3 <3 (4. . The integral is thus of order unity.7) give the width of the string. The compton wave lengths of these two mesons X = m. In order that the strings are really strings. According to the paper by Goldstone. we must have the penetration depth X and the coherence length £ small compared to the characteristic length.1) by calculating the energydensity at rest for the vortex solution given in sect. * A ctrincr i« hv definition a thin nracticallv one dimensional structure.
(4. What is important.12) (4. dual strings. however. i.13) (4. P. and so we would like to postulate that if nature were to be described by the LandauGinzburglike model of ours. is just that it is of order unity. so that we may have a pure dual string theory in the low energy range of this strong coupling limit of the LandauGinzburglike theory.54 H. We may hope that a third kind of excitation is not going to be important at low energies. it would have chosen . Olesen. in which the two coupling constants e and c^ are infinitely large.15) (4. We thus would like to suggest that such a strong coupling limit is the one in which a dual string theory emerges. being the lower quantum mechanical levels of the dual string. The order of magnitude of the characteristic length for the hadrons. In order that thin strings are a good approximation. Nielsen.e. These requirements might also be written mv. we thus need to have s/a~>\. and not by the fundamental vector and scalar particles in the theory revealed by Higgs. Thus in this limit low energy phenomena (low energy meaning energies of the order of 1/Va7) should be dominated by hadrons.14) (4.B. which again implies ^~e>\ and ^ * > ^ > 1 .16) y/a which means that the particles corresponding directly to the local fields have masses my and m s much larger than the typical hadron masses.ms>—. is thus v?^ y ^ . Dual strings A crude estimate of the energy per unit length of the string due to the core where 101 no longer has its vacuum value also gives something of the order of magnitude since c 2 0o *S t n e energy density.$. Thus we can conclude that the energydensity along the string to be identified with \\2na In the dual model is 1 i c2 7na 4 The exact ratio a'c 2 /c 4 can be computed numerically by solving the differential equations.
We have no real solution to offer to this problem. since from (4. 5. X l ' '  1 2 2 L <\. with a length of the order of magnitude of its width. Dual strings 55 fairly large values of e and v c 4 . classical field considerations very doubtful. The problem of quantization We now mention a few words concerning the problem of quantizing the above string scheme.A <» e e'V*. 2 by estimating the fluctuations in a coherent state simulating our classical solution. using eq. \/c^ > °° the typical energy is much smaller than the typical frequency. because it is in a certain sense a super quantummechanical limit.H. Olesen. (4.17) a' in the strong coupling limit. That means that the typical action of the theory is.14) and (4. make however. Thus we have an extreme quantum mechanical problem. where the coeffi= cients e and eM are small quantities. So in our limit e. i. Thus we may be in serious trouble in going from the classical theory to the quantized theory. The limit e.14). since in a classical theory the typical action. if one calculates the characteristic action of the present theory. is very large compared to h. The only thing .18) a my a mt e so that from the point of view of strong coupling theory in this limit the quantum of action h = 1 is tremendously big compared to the typical amount of action e~^. we may be in a very bad shape. But the typical energy of the field theory is rather the energy of say a vortex line. (4. The extreme quantum mechanical nature of the theory in the strong coupling limit (e. VC4 * °° is a super quantum mechanical one. 4.B. sec. say 1 erg. P. Nielsen. This is at first unclear in our formulation. So the theory is very very far from being classical. so that we could get pure dual string properties of the theory at low energies. \fc^ large.my (4. then it turns out to be much larger than Planck's constant. \/c^ *• °°) has the implication that if we want to justify our classical solution of the fields around and in a vortex line in sect.e. because we used the quantum of action as a unit# = 1. To see that our limit e.6) and (4. and that is of the order of magnitude A/a'. the classical solution of the equations revealing the particles as solutions 4>(x) «* " e~*Px .16) ^<ms. As mentioned near the end of sect. something like •H*°°. we may just remark that the masses w v and ms of the fundamental particles of the field theory are typical harmonic oscillator frequencies. where one might expect important fluctuations in a coherent state approximating our classical vortex solution.
it may be that the solution to this apparent paradox is that the positive definite character of a classical Hamiltonian may not carry over into quantum field theory.18) is correct. and it appears as a mystery how it can generate a nondefinite spectrum in the quantized version.B. that would all lead to the simple Nambu string with no extra degrees of freedom in the strong coupling limit at low energies. and we should also obtain a tachyon. This has not yet been done. We do. in which case we should obtain the usual quantized dual string. Conclusions We have seen that it is possible to make field theories that (classically) have solutions corresponding to vortexlines that are one dimensional structures moving around e. A recent discussion of this possibility has been given by 't Hooft [7]. This limiting behaviour would of course not be correct if we had other stable solutions in addition to the vortex solution. Dual strings we can say at present is the following: suppose that one could show that in the strong coupling limit the classical theory has only one stable solution. However. one of the lowest hadron states) should be much larger than the width of it. the classical Hamiltonian is certainly positive definite. In order that the length of the string in one of the lowest mass eigenstates (i. it is not at all guaranteed that the signs of the renormahzed couplings are the same as the signs of the bare couplings. c4 *• °° in the GinzburgLandau model used as the example. Many models could easily be proposed. and hence positive definiteness may impose different conditions on the couplings in a classical theory (where the bare couplings enter) and in a qunatized theory (where the renormahzed coupling enter). we had to choose a strong coupling limit e. For example. The point is that we have formulated our theory in four dimensions. . however.e. The arguments in sect. suppose that the limit (5. and hence we should run into trouble with respect to quantization*. P. Nielsen. Olesen. 6.g. For Notice that in order to formulate the theory in an arbitrary number of dimensions. Then we could quantize the theory just by quantizing the Nambu Lagrangian. namely the vortex solution. not yet know to what extent the strong coupling limit (the vortex solution) should be renormahzed. if the theory has to be renormahzed. It should be stressed that the GinzburgLandau model is only an example.e. and in spite of various attempts we have not succeeded in convincing ourselves that only the vortex solution is stable (classically).56 H. i. However. 3 then shows that the Nambu Lagrangian comes out from the field theory Lagrangian. Actually the somewhat optimistic remarks above are dubious. according to the equation of motion of the Nambu dual string.. However. we have to generalize the Higgs mechanism to an arbitrary number of dimensions.
which have e. \d dx I so that strings that are narrow (vortex lines) compared to the hadronic length y/a'** dS c~* are obtained in the strong coupling (and super quantum mechanical) limit my/oT^ c~i are obtained in the strong coupling (and super quantum mechanical) vortex lines move as Nambu dual strings. which has a width of the order of magnitude \/\/cd. We believe. the SineGordonequation.B. Olesen. The energy density along the vortex line is = f 2™' J 4 9 '2 {— — arctan exp (\/cdx)\ + 2c sin (2 arctan exp (\Jcdx) dx^s/c/d. We have shown the equivalence of string models with a certain set of solutions of some field theories in a classical approximation. One is also not restricted to work in 3 +1 dimensions. In fact. giving the equation of motion 9M 9M^(x) + cd sin (d<p(x)) = 0 . i. It is readily seen that this theory classically allows for a static solution 4 _ <p(x)= — arctan exp (sjcdxj) describing a vortex line along the Xj axis. Such theories could be built in higher dimensions too. this width is equal to the Compton wavelength for the particle of the theory. Analogous to the case of the GinzburgLandau theory. probably the simplest nontrivial vortexline model is the socalled SineGordon theory in 2 + 1 dimensions having the Lagrangian £ = i(9^(*))2+Ccos(^(x)). The field <p(x) changes by 2n/d across the vortex line.e. without giving further vortexlines so that the vortexlines would still behave like Nambu dual strings. This hope cannot be taken to be true for all theories having vortex lines. since theories can be cooked up. Nielsen. That we have to take a super quantum mechanical limit necessitates that. namely the vortex lines. which does not lead to additional degrees of freedom. zerodimensional structures in excess of the one dimensional one. P. Also. (i) the .H. Dual strings 57 instance it is easy to add some extra fields giving particles with high masses of the order of magnitude of w v and w s to the GinzburgLandau theory discussed above. That is to say one could make a fieldtheory model which also had kink like type of singularity similar to the solution in the 1 +1 dimensional SineGordon theory. but we have not proven that in the strong coupling limit at low energies all states of the GinzburgLandau field theory are states that can be described as states of some system of strings. as is seen by considering the KleinGordon equation with mass square m 2 = cd2 obtained as the weak field limit of the SineGordonequation..g. in the appendix we discuss the case of YangMills fields.
in a low energy and strong coupling limit). Dual strings field theories to be used should be quantized and (ii) the equivalence of the field theory and the corresponding string theory be proven on a quantum mechanical level (that can only be done.g. Discussion of the YangMills type of model It is natural to ask whether it is possible to produce strings from Lagrangian having internal degrees of freedom. P. Further. If we can manage to keep the vortex solutions for such Lagrangians. We also thank B. (A. is the YangMills type of Lagrangian. as is easily seen to be the case for the HiggsGinzburgLandau model. Also we have to understand how it can happen that tachyons appear in a model having at first a positive definite Harniltonian density.3) . First of all we want to thank Don Weingarten for pointing out how to make vortexline SineGordon models in any dimensions. since the string would then carry internal degrees of freedom. Olesen. Appendix. provided we could solve the problem of colliding strings). and C. An example of a Lagrangian with internal degrees of freedom which immediately comes to the mind. Tze for finding literature. but we have not even made the 26dimensional classical GinzburgLandau model yet. Here we introduce a field B nv> B =9 B . and would therefore perhaps lead to a more realistic spectrum of hadrons (and perhaps to a more realistic dual amplitude. we should like to know what the sifnificance of the critical dimension (d = 26 in the conventional model) is in terms of fieldtheory models like the ones discussed.H. the possibility of constructing fieldtheory models for strings like the ones we discussed. (A. might be an easier way to come across a good (possibly unitary) Venezianomodel than to make a string model directly. isospin or SU(3). unfortunately. Secondly we want to thank our colleagues at the Niels Bohr Institute and CERN for helpful'discussions .9 5 gB XB . Nielsen.l) Defining the dual field it is easily seen that ( 3 " + * 2 * " X ) 5 * „ = 0. if at all.58 H. this would indeed be very nice from the point of view of the dual string. e. does not seem to be the case.B. Zumino for calling our attention to the fact that the equation we discussed is well known as the GinzburgLandau equation. This. If some day one understands the quantum properties better.
we obtain from the Lagrangian (A.f B do»v ./„ = 0 for a free YangMills field. Nielsen. (A. and hence we have the interpretation that the flux is given by < = JB D do*" (static case) . The reader may wonder why we intro} duce two fields 4 and v/ and not just one field.8) by.9) i. namely £=&llvB>l''+{[(dll+gBllX)+]2+±[(<ltt+gBllX) + c24»2 . a similar procedure leads to B J B»XB* dVv = JB da"" .bvBfi in B^v for large distances . If. and 5 o k = 0.7) in a way similar to the Higgs (GinzburgLandau) Lagrangian discussed in sect.10) Considering now the static solution with cylindrical symmetry we see that the term Bu X Bv in eq.5) S where V is a volume bounded by the two surfaces Sj and S2. Dual strings which is the analogue of the Maxwell equation 59 3MC =0' <A4> which states that magnetic monopoles do not exist.. The reason for this will turn out > later. P. varying B^ Jv=g(4>Xdv<l>)+gWXdJ)+g2(BvX4>)X4>+g2(BvX*)X* . we can take B0 = 0. Now we are looking for a solution which quantizes the flux (A.H.3) is concerned.3) poses a problem with respect to the interpretation of (magnetic) flux lines. Eq. we concentrate on the static situation. at least in general flux lines can originate inside the volume. Olesen.4) one can derive that 0= f b^F* dV= »i f F l da^ . As far as eq. The form of eq. From eq. however.c4(4>2)2 +d2*2 d4W2)2 + }2 (A.5) tells us that no magnetic flux lines can start inside the volume. (A.8) +e2<j>4/ e 4 (<j>*) 2 .6) Thus.e.7) Next let us consider a specific YangMills Lagrangian. (A.B. (A.f F S 2 do"" . (A. (A. where we shall see that in order to have a vortex solution at least two isovector fields are needed. (A. (A. (A.l) is smaller than the term b Bv . (A. Defining the current to be 3 M fl liv +gB tiv XB^=j v . where the fields <> and \\i are isovector fields. 2.
Notice that due to the last condition (A. (A. in the vacuum the lenghts of the isovectors <> and \/ are fixed. Now the ground state (the vacuum) of the Lagrangian (A. This corresponds to selecting the flux lines in the 3direction.g. Olesen. V g V*> (A15) where x is the phase of 0j + z'02. It is easily seen that the condition (A. This ensures that in any frame of reference in isospace at least three components of <> and \/ are nonvanishing (e. and eq.12) +2=t4' *2=^> ** = ^ > (A13) where we assume that 4> and ty are not in the same (or opposite) direction. (A. and in addition the projection of one vector on the other is fixed. c4d4 Hi) • Vc4 < i) A4  Thus.B. This result only follows if we have at least two fields < and \». Inserting eq.16) i. (A. J > Having obtained the flux quantization we then go to the strong coupling limit.8) corresponds to (A.e. in order that the width of the vortex line is made sufficiently small. .. large distance).317 60 H. i.X<>)X<}> +(*„Xi>)X«» =(fi>)<t>+(^v/)»/(<t) 2 +vl» 2 )^ =j(4>xa„4»+ *X3„«»). (A. 0 ! . Dual strings from the axis of symmetry. (A..l 1) provided we integrate over a circle with large radius.12) for the current to vanish at large distances imposes the condition on Bv that Bl=B2 = 0. Thus we can write the flux (A. The width is given We have not convinced ourselves that there are no other quantas than 2n/g.. the flux is quantized*. 3 .7) as 0 = (j)B 0)djc M (static case. Nielsen. vanishes. 10) leads to (5.15) in eq. For such large distances/. P.13) the phase of <//j +i^2 ' s the same asx + constant. i//j and i// 2 ) Now let us go to a frame of reference where 0 3 = 1//3 = 0.l 1) and using the fact that the phase is only unique modulus 27W we get that the flux is given by I 4> I = 27TH/S .e.
[8] for example. As long as log (A/£) is not too large. From the very general argument in sect. Sarma and E. [7] G. "t Hooft. (6J D. on high energy physics. References [1] Y. (Wayne State University.) 2 (5 M «j/) 2 ] . . Phys. (8] A. Thomas. Canberra preprint (1973).B. J. m's = \/2d2 .[T2 W + d^ (A 19) V=* K ^ 2 V ' Proceeding as in sect. 15th Int.H. on symmetries and quark models.e. Nuovo Cimento. Conf.17) The Higgs vector particle is obtained from the seagull terms in the Lagrangian (A. A preprint of L. [5] H.J.B. Fassie [9] with a similar idea has appeared. See ref.J. which then gives us the string solution. Niels Bohr Institute preprint NBIHE739. B57 (1973) 367. 1969). one can easily see that due to gauge invariance no new degrees of freedom are introduced in the string Lagrangian. H. Lines of quantized magnetic flux and the relativistic string of the dual resonance model of hadrons. we know that this solution corresponds to the classical Nambu Lagrangian.B. Dual strings 61 by the order of magnitude of the Compton wave lengths of the Higgs particles. (4) P. ^4 "* °° (or m s . PHys. Olesen. Nambu. Nielsen. Conf. SaintJames. Note added in proof In addition to the term (4. Proc. 4 we can now go to the strong coupling limit g » °°. 13) it is easily seen that all vector mesons B acquire a mass. Susskind. Olesen.9) the magnetic energydensity also contains a term coming from the seagull term in the Lagrangian. this term does not change any of the conclusions in sect. (A. P. nij. . c 4 »• °°.J. Nielsen and P. Type II superconductivity (Pergamon Press. 1970.  2  Y. Nucl. Nuovo Cimento 69A (1970) 457. G. Wy > lly/a'). from U 2 ( 5 M X * ) 2 + ^ 2 ( 5 M X i/) 2 = ^ 2 [52(d>2 + u> 2 )(/? /i <. Phys. 3. L.B. D5 (1972) 2535.N. Int. to be published. Goldstone. B56 (1973) 109. Olesen. P. Goddard.18) Inserting the vacuum values (A. As before. Superconductivity of metals and alloys (1966). Kiev. The latter term gives rise to an energydensity of the order of magnitude (c2/c4) log (A/?). we have scalar particles with masses ms=\/2c^. i. However. (A.8). Thorn. Nambu. Rebbi and C. Fassie. Mansouri. Chang and J. Nucl. 1970. Rev. De Gennes. 19) L. Lectures at the Copenhagen Summer Symposium. Nielsen. 4. C. CERN preprint TH 1666 (1973). [3] L. 1969).
. Schwinger suggested that the same effect could occur in four dimensions for sufficiently large couplings. preserving exact gauge invariance and treating the gauge fields as angular variables (which makes a gaugefixing term unnecessary). The problem is that quarks have not been seen. and Susskind4 has shown that the asymptotic states of the model contain only massive photons. There is unfortunately no Lorentz (or Euclidean) invariance in the strongcoupling limit. The polarization effects which prevent the appearance of electrons in the final state take place on a longer time scale (longer than \/my. However. where my is the photon mass).8 A brief discussion of the . The mechanism applies to gauge theories only. as Casher etal. cannot appear as separate particles in a final state. not electrons. Cornell University. have shown in detail. This structure is reminiscent of relativistic string models of hadrons. A number of speculations have been offered as to how this might happen. where the photon acquires a mass . The use of a Euclidean space (i. The lattice gauge theory has a computable strongcoupling limit. the model discussed here has a builtin ultraviolet cutoff.1 Independently of the quark problem. A new mechanism which keeps quarks bound will be proposed in this paper. the energy eigenstates (including scattering states) of the lattice theory can be determined from the "transfermatrix" formalism as has been discussed by suri 5 and reviewed by Wilson and Kogut. However. INTRODUCTION The success of the quarkconstituent picture both for resonances and for deepinelastic electron and neutrino processes makes it difficult to believe quarks do not exist. the electrons are present in deepinelastic processes and behave like free pointlike particles over short times and short distances.e. The model discussed in this paper is a gauge theory set up pn a fourdimensional Euclidean lattice. The strongcoupling expansion involves sums over all quark paths and sums over all surfaces (on the lattice) joining quark paths. The mechanism will be illustrated using the strongcoupling limit of a gauge theory in fourdimensional spacetime. for some reason. Further study of the Schwinger model by Lowenstein and Swieca3 and Casher. The confinement mechanism proposed here is soft (longtime scale). The inverse of the lattice spacing a serves as an ultraviolet cutoff. Nevertheless. Kogut. in consequence the theory is far from covariant. Schwinger observed many years ago2 that the vector mesons of a gauge theory can have a nonzero mass if vacuum polarization totally screens the charges in a gauge theory. I. Schwinger illustrated this result with the exact solution of quantum electrodynamics in one space and one time dimension. in this limit the binding mechanism applies and there are no free quarks. and in the strongcoupling limit all particle masses (including the gauge field masses) are much larger than the cutoff. Ithaca. It is shown how to quantize a gauge field theory on a discrete lattice in Euclidean spacetime. in the model discussed here the cutoff spoils the possibility of free pointlike behavior for the quarks. New York 14850 (Received 12 June 1974) A mechanism for total confinement of quarks.319 YSICAL REVIEW D V O L U M E 1 0 . similar to that of Schwinger.e 2 for any nonzero charge e [e has dimensions of (mass) 1 '' 2 in this theory]. This suggests that quarks. imaginary instead of real times) instead of a Lorentz space is not a serious r e striction. is defined which requires the existence of Abelian or nonAbelian gauge fields. NUMBER 8 15 O C T O B E R 1974 Confinement of quarks* Kenneth G. Wilson Laboratory of Nuclear Studies.
In weak coupling the gauge field behaves like a n o r m a l free z e r o . for e > ec the photon m a s s would be nonzero and v a r y with e. VI t h e r e is a brief d i s c u s s i o n of the p r o b lem of L o r e n t z invariance and the r e l a t i o n to string models. Consider. IV the strongcoupling expansion for the lattice gauge theory is explained.) In the F e y n m a n p a t h .i n t e g r a l p i c t u r e .p a r t o n model. M is nonzero and a function of t e m p e r a t u r e . In Schwinger's speculations about four d i m e n s i o n s . Speculative plot of photon mass vs renormalized charge e. T h e r e is again a question whether this change of p h a s e is f i r s t . 2. The point ec is a point of nonanalyticity. Below Tc.1) whose F o u r i e r t r a n s f o r m d e t e r m i n e s the e . the spontaneous magnetization M is 0. In Sec. in unknown units.s t a t e situation one can think of the t r a n s i t i o n from z e r o to n o n z e r o photon m a s s a s a change of p h a s e : this analogy is b e s t understood by imagining the p a r t i c l e s of quantum e l e c t r o d y n a m i c s to be the excitations of a medium (the e t h e r ) .i n t e g r a l f r a m e w o r k will be used in an intuitive r a t h e r than a f o r m a l way. called c r i t i c a l points. In s t r o n g coupling the gauge field i s m a s s i v e and the q u a r k s a r e bound. it is not known at p r e s e n t whether these difficulties can be o v e r come.0 a s T~TC from e i t h e r s i d e . Thus t h e r e should be a p h a s e t r a n s i t i o n a t s o m e i n t e r m e d i a t e value of g. By analogy with the s o l i d . In this c a s e it is the ether which u n d e r g o e s a change of p h a s e at ec. for i n stance. Speculative plot of photon mass vs renormalized charge e if there is a firstorder transition at ec . 1). In Sec. (The r e s t r i c t i o n to one field is only for simplicity.m a t r i x method i s given in Sec. Nothing is known about this t r a n s i t i o n at the p r e s e n t t i m e .320 2446 KENNETH G. III. QUARK BINDING MECHANISM The binding m e c h a n i s m will be explained in this s e c t i o n using the F e y n m a n p a t h . Consider the currentcurrent propagator £>„„(*) = <nTJ„(x)J„(0)n> + (2. a f e r r o m a g n e t in the absence of an e x t e r n a l field. occur in s o l i d . F i g u r e 1 shows how a plot of my vs e might look. A s s u m e that the q u a r k s i n t e r a c t through a single gauge field.) The model d i s c u s s e d in this p a p e r i s a single Abelian gauge field coupled (with strength g) to m a s s i v e q u a r k s . II the n a t u r e of the q u a r k binding m e c h a n i s m will be d i s c u s s e d .o r d e r p h a s e t r a n s i tion (critical point) for which M . An e x t r a o r d i n a r y feature of the strongcoupling expansion of the lattice theory (see Sec. Ill the gauge theory will be formulated on a d i s c r e t e lattice. (Coleman and Weinberg 7 have found a nont r i v i a l example of a f i r s t . the s u r f a c e s a r e generated by the gauge field t r e a t e d in s t r o n g coupling. In Sec. T h e r e a r e g e o m e t r i c a l difficulties in r e l a t  e— FIG. WILSON 10 t r a n s f e r . This will be explained briefly in Sec.o r d e r (cf.s t a t e physics at c e r t a i n types of phase t r a n s i t i o n s . F o r any t e m p e r a t u r e above the Curie t e m p e r a t u r e Tc. 2) or s e c o n d . The paths and s u r f a c e s a r e defined on a d i s c r e t e l a t tice. 1. The p a t h . In Sec. both c l a s s i c a l l y and quantum m e c h anically.o r d e r t r a n s i t i o n in a n other context. The transition at ec is secondorder (see text). II. In Sec. III.i n t e g r a l p i c t u r e the p r o p a g a t o r FIG. . At T c t h e r e may be e i t h e r a f i r s t o r d e r phase t r a n s i t i o n (in which c a s e M is d i s continuous a t Tc) o r a s e c o n d .o r d e r (Fig.e " annihilation c r o s s section into h a d r o n s . V a c u r s o r y d i s c u s s i o n of weak coupling i s given. Fig. Similar nonanalytic points.m a s s field (despite modifications introduced in the lattice quantization) and the q u a r k s a r e unbound. IV) is that it has the s a m e g e n e r a l s t r u c t u r e a s the r e l a t i v i s tic s t r i n g models of h a d r o n s . The quantization p r o c e d u r e and strongcoupling approximation d e s c r i b e d in this paper can be a p plied to nonAbelian gauge t h e o r i e s a l s o . A s s u m e that the c u r r e n t s Ju{x) a r e built from q u a r k fields as in the q u a r k . 8 The vacuum e x pectation v a l u e s of the gauge theory involve (in the strongcoupling expansion) s u m s over all quark paths and s u m s over all s u r f a c e s connecting these p a t h s . the photon m a s s would be z e r o for any charge e l e s s than a c r i t i c a l coupling e c . showing the existence of the second p h a s e . qualitatively. ing the s u r f a c e s on the lattice to the continuum s u r f a c e s of the s t r i n g m o d e l s .
the important paths in the Feynman path integral bear no detailed relation to possible physical final states (the paths are paths of bare particles. 5. including the loop joining the points 0 and x. a more difficult question to answer and will not be discussed in this paper. 6(b)]. This is FIG. where A^x) is the gauge field.(0) are thought of as producing a quark pair at the origin which later annihilate at the point x: One has to sum over all paths joining the points 0 and x for each of the pair of quarks. Example of current loop (as In Fig. (Integration over the gauge field produces gauge propagators which connect these loops. atleast when x and 0 are far apart. where m is the quark mass. 3) with an extra vacuum loop. Suppose the gaugefield averaging is performed before the quarkpaths averaging. (Here . independently of the quark paths there is another weight factor. If vacuum loops are important enough then space will be filled with a high density of vacuumproduced quarkantiquark pairs. An example of quark (q) and antiquark (q) paths connecting the points 0 and x. Note that x must be large: If x is small there is little likelihood of finding a large size loop. First. It is also possible to have independent loops for the points 0 and x (Fig.) . 5). The question then is whether there can be an excess of quarks over antiquarks. this leads to further closed loops as illustrated in Fig. The currents «/„(*) and </i. FIG. 4. large x does not necessarily mean small q. Finally. 3. Then one determine the average over all gauge fields of the weight factor exp[ifgAll(y)ds1'] weighted further with the exponential of the free gaugefield action. in a region of macroscopic size. not large x.means up to a power of x2. not physical particles). For an Abelian gauge theory this average can be computed explicitly: It is A o 0. This is illustrated in Fig. It is assumed in this discussion that vacuum loops are not important. FIG. then clearly no detector will see a quark or antiquark in isolation. the exponential of the free action for the gauge field. say 10"13 cm [see Fig. there will be many quarkantiquark pairs inside a detector of macroscopic size. 6(a). The answer to this paradox has two parts. 3. In fact the study of whether wellseparated quarkantiquark paths exist for large x is really a search for a quarkantiquark threshold in e*e~ annihilation. The weight associated with a given quark path or set of paths includes a factor of expligfA^x) xds11]. All possible loops must be summed over too. Here f • • •ds' i is a line integral or a sum over line integrals for each of the quarkantiquark loops. In particular. Example of separate quark loops for the points 0 and x.321 C O N F I N E M E N T O F QUARKS 2447 10 D v(x) is given by a weighted average over all possible classical quark paths and all possible classical values of the gauge field. There are further weight factors independent of A. Secondly. If the quark and antiquark paths are unlikely to separate beyond a fixed size. The vacuum can also emit and absorb quark pairs. In order that quarks exist as separate finalstate particles it must be possible to have quarkantiquark loops with wellseparated quark and antiquark lines. The constant g is the coupling constant of the gauge theory. 4. namely. which would contribute a term ~exp(2miSx2~) to the currentcurrent propagator for large x. The combined weight factor is then averaged over all quark paths and all gauge fields A^x) to give the currentcurrent propagator. This may seem a bit peculiar: One expects quarks to appear in the final state of e* e~ annihilation only at large virtualphoton momentum q if they appear at all.) Such a term corresponds in momentum space to the singularity at the threshold q2 = (2mf. or vice versa. but this possibility will not be important here. An example of paths for the quark and antiquark are shown in Fig. and large q means small x.
322
2448 KENNETH G. WILSON 10
exp
g2j> ds^ds^D^y
>•')]
w h e r e D^v{y y') is the free gaugefield p r o p a gator. The quark binding m e c h a n i s m can be s e e n by comparing the above e x p r e s s i o n for one s p a c e dimension and t h r e e space d i m e n s i o n s . In t h r e e space dimensions this calculation gives no binding (the binding o c c u r s only with a modified gauge field action: s e e Sec. IV), while t h e r e i s binding in one space dimension. In t h r e e s p a c e d i m e n sions D^uiy y') b e h a v e s a s (y y'Y2. In consequence l a r g e values of (y'  v) a r e negligible in the double line i n t e g r a l . Hence, the double i n t e g r a l is p r o p o r t i o n a l to P , where P is the length of the loop. Unfortunately the integral is divergent at y' =y; a cutoff is needed for the integral to make s e n s e . Since a cutoff will be introduced a n y way in this paper, this i s not a major concern. F o r simple loops, the p e r i m e t e r P i s roughly of o r d e r {x2)^2 (ignoring the c a s e that x is close to the light cone). Thus one h a s an exponential of the type one expects when free q u a r k s a r e p r e s e n t . In one s p a c e dimension, D^u(y  > ' ' ) behaves a s lnf(y y')2] for y' y l a r g e , and y' and y can freely range s e p a r a t e l y over the loop. In this c a s e the double i n t e g r a l is p r o p o r t i o n a l to P2. Now the gaugefield a v e r a g e behaves as e' ic.p2 w h e r e c is a constant. In this c a s e the contribution of l a r g e loops is heavily s u p p r e s s e d and t h e r e a r e no free q u a r k s . [The c a s e of nearby q u a r k  a n t i q u a r k p a i r s as in Fig. 6(b) is special—in this c a s e l a r g e y  y ' is unimportant due to cancellation between the quark path and the nearby and oppositely d i r e c t e d antiquark path. In this c a s e the double i n t e g r a l behaves as P , not P2, but in this c a s e t h e r e a r e no isolated q u a r k s . ] In the strongly coupled lattice gauge theory d e s c r i b e d in l a t e r s e c t i o n s , the gaugefield a v e r a g e of exp[igiAu(x)ds1'] behaves a s exp(ic'A), where A i s the enclosed area of the loop. This heavily s u p p r e s s e s l a r g e loops, such as in Fig. 6(a), w h e r e A is of o r d e r P2. One can think of one factor P a s being roughly (* 2 ) l / 2 , the other P a s being a n a l o gous to a m a s s multiplying ( x 2 ) l / 2 . Since P~ » a s x  ° o , the q u a r k  a n t i q u a r k threshold is at infinite mass. In all these calculations one can have a l a r g e loop if t h e r e i s a nearby vacuum loop (Fig. 7). In this case one always gets e'c p behavior. F o r e x ample, in the strongcoupling c a s e the r e l e v a n t enclosed a r e a i s the a r e a between the two loops which is p r o p o r t i o n a l to the p e r i m e t e r P provided the s e p a r a t i o n of the two loops is fixed independently of P. This is in accord with Schwinger's p i c t u r e . While an isolated w e l l  s e p a r a t e d loop
FIG. 6. (a) Loop with wellseparated quark and antiquark. (b) Loop with small separation between quark and antiquark. may be s u p p r e s s e d (due to P 2 or A dependence in the exponential) a loop closely shielded by a v a c u u m  p o l a r i z a t i o n loop is always u n s u p p r e s s e d . The binding m e c h a n i s m proposed h e r e i s soft: The exponential damping i s a s s o c i a t e d with l a r g e s i z e loops having l a r g e a r e a s . The behavior of s m a l l loops is i r r e l e v a n t to the binding m e c h a n i s m . Also for s m a l l loops both their a r e a and p e r i m e t e r a r e s m a l l and neither i s of g r e a t i m p o r t a n c e in an exponential. The m e c h a n i s m d i s c u s s e d h e r e w o r k s equally well for Dirac q u a r k s or s c a l a r q u a r k s . This i s in c o n t r a s t to the Higgs m e c h a n i s m which can wipe out the charge of s c a l a r p a r t i c l e s only. HI. LATTICE QUANTIZATION OF GAUGE FIELDS
A. Classical action on a lattice
In this section the gaugefield action (space t i m e integral of the L a g r a n g i a n ) will be defined on a d i s c r e t e lattice with spacing a in Euclidean s p a c e  t i m e . The s i m p l e s t way to p r o c e e d i s to consider a continuum action, substitute f i n i t e difference approximations for d e r i v a t i v e s , and r e p l a c e the s p a c e  t i m e i n t e g r a l by a s u m over the lattice s i t e s . However, the r e s u l t of this is an action which i s not g a u g e  i n v a r i a n t for n o n z e r o a. Because of the v a g a r i e s of r e n o r m a l i z a t i o n this is likely to mean that the quantized theory still lacks
FIG. 7. Quarkantiquark loop with nearby vacuum loop.
323
10 C O N F I N E M E N T O F QUARKS
••agAn
2449
(3.6)
gauge invariance in the limit a—0 (if such a limit exists). The alternative is to formulate gauge invariance for a lattice theory, and tinker with the action so that it is gauge invariant for any a. This alternative will be pursued here. For convenience a charged Dirac field <l> coupled to a single gauge field Ap will be discussed in detail. Generalizations to nonAbelian gauge groups will be noted later. On the lattice the fields are iji„, $„,. and Anfl, where n is a fourvector with integer components referring to points on a simple hypercubic lattice. A simple action on the lattice for the Dirac field (ignoring the gauge field for now) has the form
Ai,= a Yim0il>„il>„
i
In the action At, the field Bn M acts like an angular variable: A$ is periodic in Bnll with period 2TT. The free gaugefield action will be defined to preserve this property. This does not mean that A„ M is an angular variable in the continuum limit. Owing to the relation (3.6), Anll becomes an infinitesimal angular variable A^ix) for a —0; such a variable has the range <*><All(x)<«> without any periodicity. A gaugeinvariant lattice approximation for VpAuVyAn is
F„jiu\An.
AnVAvts.n+Anll)/a
(3.7)
It is convenient to define a rescaled form of F„ • nil" (3.1)
fnlivz
<l
M
*&*
 BnU + B „ + ( j i  /  **n+u.l
where mQ is the bare mass; A is a unit vector along the axis \i; ( i 4 ^ , replaces the spacetime integration of the continuum theory, and {ipn+ j _^ n _j)/2a replaces V^. There is no overall factor of i due to the Euclidean metric. A gauge transformation on the lattice can be defined as follows;:
(3.2) (3.3) (3.4)
(3.8)
A simple lattice action for the gauge field which preserves periodicity is 1
2p
e "nM"
(3.9)
In the continuum limit, f„liv — 0 due to the factor a2 in the definition oifnllv. Thus for small a, one can write
* •  *
£
tt+ifn»vifnltU2)
(3.10)
where g is the coupling constant and yn is arbitrary. The terms ifntn+U an( * 'I'n'Pni' a r e n 0 * i n " variant to this transformation; the corresponding gaugeinvariant expressions are ^„^n+5 xexp(iagAnli) and $„!/<„;; exp(jagvl„_p .„). Thus, a gaugeinvariant form for \ is
The constant term is irrelevant. The linear term in/„,,!, is 0 because / B(1 „ is odd in the indices fi and v. The quadratic term gives Ag** —ter 2_, Fnliv , (3.11)
 « « £ «<,?„#•, •
n
(3.5)
It is convenient to define
A=cJ2$ntn+KT,12®«r^fieiB"»y»+nyi'4>neiBnl>)+~J2
which is the conventional gaugefield action in a lattice approximation. The terms involving /n M«3> fn \iv*> e t c  a 1 1 vanish for a— 0 even after removing a factor a4 to convert Yjn i n *° an integral. The full action may now be written
£ «"M» ,
2g*
(3.12)
with c = m0a4, K=a3/2. This action reduces to the usual continuum action for a — 0; for finite a it is gauge invariant and periodic in the gauge field. Note, however, that the continuum limit is a classical limit in which the lattice variables #„., ?„, and An)1 approach continuum functions ip(x), ?(*), and A„(x) with x = na. The continuum limit of the quantized theory is much harder to discuss owing to renormalization problems.
B. Quantization
The problem of principal interest here is the quantization of the gauge field. Therefore, the gauge field will be quantized by itself to start with. Later the quantization of ip will be discussed. At the end of this section the generalizations to nonAbelian gauge theories will also be described. The quantization of the lattice gauge theory will
324
2450 KENNETH G. WILSON 10
be c a r r i e d out in two s t e p s . The f i r s t step will be to define a lattice v e r s i o n of Euclidean vacuum expectation v a l u e s , s t a r t i n g from a lattice v e r sion of the F e y n m a n path i n t e g r a l . The second s t e p will be to define a quantum theory on the l a t tice, which will allow the introduction of a r e a l time v a r i a b l e and the definition of p a r t i c l e s t a t e s and s c a t t e r i n g a m p l i t u d e s . In both c a s e s the l a t tice p r o v i d e s an u l t r a v i o l e t cutoff and t h e r e is no L o r e n t z i n v a r i a n c e . L o r e n t z invariance can only be achieved in the l i m i t a (lattice spacing) —0, if at all, and in p r a c t i c e this is a difficult limit to evaluate. As d i s c u s s e d in Sec. I, one would like to c a l c u late the gaugefield a v e r a g e of exp[igfAll(x)ds1'] weighted with the gaugefield action. On a lattice the line i n t e g r a l b e c o m e s a s u m over a closed path P on the lattice (see, e.g., Fig. 8). The s u m has the f o r m : i^p(,±)Bnli, where a p a r t i c u l a r Bnll i s p r e s e n t in the s u m if the path connects the s i t e s n and n + jl {BnU a p p e a r s if the path goes from n + jlto n). On the l a t t i c e , an a v e r a g e over all gauge fields involves i n t e g r a t i n g over all values of the Bnll for all n and \i. Normally one would have i n t e g r a l s over an infinite r a n g e : °°<Bnll <°°, but because of the periodicity in Bnll t h e r e i s no point to i n tegrating over m o r e than a single p e r i o d . Thus the lattice v e r s i o n of the gaugefield average is 1{P) = Z
(0,0) (1,0)
—)
P +
(2,0)
FIG. 8. Example of a lattice path P. finite lattice volume), but these d i v e r g e n c e s a r e normally r e m o v e d by the division by Z (this division is equivalent to r e m o v i n g all vacuum loops in p e r t u r b a t i o n theory). One can define m o r e conventional vacuum e x pectation values in a s i m i l a r m a n n e r . F o r i n stance, one can define a p r o p a g a t o r . In the a b sence of a gaugefixing t e r m it i s awkward to d e fine a p r o p a g a t o r for the gauge field Bnil itself; instead one can define a g a u g e  i n v a r i a n t p r o p a gator as
nnj
m V
IT
dB„ 1 2g2
Ji
xexp
(if„Mifoa
Z
eif (3.15)
'(nnf
dBm
< exp i
£>)£„,+ ^ 2g
nfju
2>'M>
(3.13)
with
nnj ' e x p i^£ e '
dB,
I 1
m v J Ti
fnfv
(3.14) Note that no gaugefixing t e r m h a s been added to the action. The finite r a n g e of Bnll m a k e s a gaugefixing t e r m u n n e c e s s a r y . In continuum gauge t h e o r i e s w h e r e A^ix) has an infinite r a n g e [«><All(x)<<*>\ a gaugefixing t e r m is e s s e n t i a l to have a convergent functional i n t e g r a l . The r e a son for this is that the volume in p a t h  i n t e g r a l s p a c e g e n e r a t e d by all possible gauge t r a n s f o r m a tions is infinite; the gaugefixing t e r m provides a convergence factor in this volume. 1 0 In the l a t tice theory the total volume of integration is finite if the lattice itself is of finite extent; no c o n v e r gence factor i s r e q u i r e d . F o r a lattice of infinite extent t h e r e a r e d i v e r g e n c e s due to the infinite number of i n t e g r a t i o n s (in other w o r d s , the i n 
This is a p r o p a g a t o r for the o p e r a t o r e 'nvv: it is defined only for the lattice points n of a Euclidean s p a c e  t i m e lattice. If the lattice spacing i s a, this m e a n s the p r o p a g a t o r is defined only for i m a g inary t i m e s of the form in0a, w h e r e n0 is an integer. A theory defined only for d i s c r e t e imaginary values of the t i m e leaves much to be d e s i r e d . Fortunately, one can g e n e r a l i z e the theory to d e fine a Hamiltonian for a quantized theory. The p a r t i c l e e i g e n s t a t e s and s c a t t e r i n g amplitudes of the theory can then be obtained, in p r i n c i p l e , by diagonalizing the Hamiltonian. The Hamiltonian will be defined using the t r a n s f e r  m a t r i x f o r m a l i s m . Only a brief d i s c u s s i o n of the t r a n s f e r  m a t r i x approach will be given h e r e . F o r a review of the ideas s e e Wilson and Kogut. 6 A detailed d i s cussion including a p p r o x i m a t e calculation of s i n g l e  p a r t i c l e e n e r g i e s and s c a t t e r i n g amplitudes in a simple s c a l a r field theory is given by s u r i . 5 Consider the e x p r e s s i o n for Z. Introduce finite bounds on the lattice coordinate n0, say
N^nn^N
(3.16)
Introduce p e r i o d i c boundary conditions (see b e low). Then one can w r i t e Z a s the t r a c e of a m a t r i x V, m o r e p r e c i s e l y
325
10
Z = TTV2N+1
CONFINEMENT (3.17)
OF QUARKS
2451
This f o r m u l a i s made p o s s i b l e by the fact that each t e r m in the action A involves no m o r e than two adjacent values of n0. To s e t up the m a t r i x V, one must f i r s t u n d e r stand the s p a c e on which i t a c t s . The space used h e r e i s the s p a c e of a l l functions tyiB^,) (periodic in each B^t with period 2n), w h e r e the index i r u n s from 1 to 3 only, and the lattice v a r i a b l e n has only t h r e e components (nlt n2, n3). The m a t r i x Vwill be defined a s a function of two s e t s of a r g u m e n t s , s a y B~i and B~t, these two s e t s of a r g u m e n t s r e f e r r i n g to the s p a c e  t i m e fields Bni for two adjacent values of n0. Matrix multiplication of two V ' s involves i n t e g r a t i o n s over a s e t of v a r i a b l e s {B»j}. Define V a s V (3.18)
of E q s . (3.18)(3.21). One m u s t r e m e m b e r that the v a r i a b l e s 5 * 0 a r e integrated out in the d e finition of V. This m e a n s that in forming the c o m plex conjugate of V one c a n also make the change of v a r i a b l e B^0~  £ * „ . ] Hence V h a s a complete orthogonal s e t of e i g e n s t a t e s * and eigenvalues A. The Hamiltonian H i s now defined a s follows: The e i g e n s t a t e s of H a r e the eigenstates. of V; the eigenvalues of H a r e given by E a'1 InA' (3.23)
The quantity f a s w r i t t e n out below looks c o m plicated, but a l l it i s i s that p a r t of the action A r e f e r r i n g to a given n e a r e s t  n e i g h b o r pair of v a l u e s for n0; t e r m s in A r e f e r r i n g to a single value of n0 a r e divided equally between the m a t r i c e s connecting n0 to n 0 + l and n0 to n0 1. The r e s u l t is
w h e r e X is the c o r r e s p o n d i n g eigenvalue of V. The r e a s o n for the factor a" 1 will be evident shortly. The r e a s o n for using the l o g a r i t h m is so that the e n e r g i e s of m u l t i p a r t i c l e s c a t t e r i n g s t a t e s will be the s u m of s i n g l e  p a r t i c l e e n e r g i e s (see suri 5 ). A p r o b l e m a r i s e s with this definition if V has any negative eigenvalues A. If this w e r e to happen, H would have complex eigenvalues. This did not happen in the c a s e studied by s u r i 5 ; whether it happens h e r e the author does not know; this q u e s tion must be studied f u r t h e r . Even if V has negative eigenvalues, they may be i r r e l e v a n t in the l i m i t a—0 if such a l i m i t e x i s t s . The definition (3.23) m e a n s that V=e(3.24)
f=iEEZ(^«"i») 4ff
2g»> where Bti and fni BZi + B^i . 0 " Effi ~ B no • (3.21)
+ ^ n + i ',: ~'
B
(3.19)
n i
n + j , i ~•BHi
(3.20)
V/ith the definition of V given h e r e , the t r a c e TrV 2 J V + 1 i s easily s e e n to r e p r o d u c e a l l the i n t e g r a t i o n s involved in the equation for Z , and the sum of the 27V+1 exponents [ / r e p r o d u c e s the a c tion A except for s o m e additional t e r m s coupling the boundary n0 = N to the boundary n0 = N to achieve a p e r i o d i c s t r u c t u r e . Note that the 0 components of Bnfl have r e c e i v e d s p e c i a l t r e a t m e n t . T h i s i s b e c a u s e t h e r e a r e no t e r m s in A involving B„0 for m o r e than one value of n0; this m a k e s it p o s s i b l e to include i n t e g r a tions over Bm in the definition of V r a t h e r than in the definition of m a t r i x multiplication. The m a t r i x F i s used to define the quantized theory. Briefly, this i s accomplished a s follows. V is a Hermitian matrix, i.e., V{B,B')*=V(B',B). (3.22)
This m e a n s V i s the o p e r a t o r which p r o p a g a t e s a state through a n imaginary time ia. It i s a consequence of this that the p r o p a g a t o r Dnll^fOT is a vacuum expectation value for imaginary time in0a in the theory with Hamiltonian H. F o r proof of this s t a t e m e n t s e e Refs. 5 and 6. The lattice quantization p r o c e d u r e can be e x tended to nonAbelian gauge t h e o r i e s . This is done a s follows. In place of the single variable Bnji, one h a s a s e t of v a r i a b l e s B"v where a i s a n internal index. F o r each n and \x, B"^ is to p a r a m e t r i z e a n element bnfl of the gauge group. In place of exp{iBnll) one s u b s t i t u t e s U(b„u), where U is the u n i t a r y m a t r i x r e p r e s e n t i n g bn)i in the quark r e p r e s e n t a t i o n . The product ij>„Y,,$„+$ exp{iBn)1) i s r e p l a c e d by ^ y ^ U{bnt)ip„+~. A gauge t r a n s f o r m a tion i s defined by a s e t of group t r a n s f o r m a t i o n s >•„; under t h e s e t r a n s f o r m a t i o n s
# „  U{yMn ,
(3.25) (3.26) (3.27)
y„tynuHy„),
U{bnu)U{yn)U(bnv)UHyBt(r)
:
(3.28)
[This r e s u l t can b e verified by close examination
w h e r e ynbnllyn+$~l i s computed according to the multiplication law of the group. A simple gaugeinvariant action for the gauge field is
326
2452
A
K E N N E T H G. WILSON B= 0T2 L Tr0 1 (6 n M 6 n + ? i „6 n + ; ; t ( i  1 6 n l /  1 ),
(3.29)
10
where the unitary transformation UA is taken from the adjoint representation of the group. (Any representation will do as well.) In the classical continuum limit, bn„ becomes an infinitesimal group transformation fS°M =gaA"(na) with A" fixed as a — Oj; in this limit one can show, with some effort, that AB reduces to the standard continuum YangMills action. When the nonAbelian theory is quantized, the integrations over bmv are group integrations over all elements of the group. These are compact integrations (for compact groups) and no gaugefixing term is required. Since quark binding can be illustrated using Abelian theory, the nonAbelian theory will not be studied further. Finally, the quantization of the Dirac field will be discussed. It is convenient initially to quantize the Dirac field in an analogous manner to the gauge field. The only problem that occurs is to define "integration" for a Fermi field. This can be done.11 The property of the integral that is crucial for quantization is translational invariance in the integration variable. For example, when quantizing a scalar field 0 on a lattice the field averaging involves the integral fZ^dtpn which has the translational invariance  ^ r f 0 „ / ( 0 „ + J„) =  y ( 0 n ) d 0 „ (3.30)
the bracket (• • •) must vanish for all products of #'s and iji's, except the product containing all possible i//s and ip's. For example, suppose there are two lattice sites 0 and 1, and ip„ and <jin are single component fields. Then the brackets must be <1)=0,
<W=<^>=<W=W=0>
( M , ) = t e ) == 0, (3.32)
(?M) = • • • = o ,
<WoWi> = 1 . where 1 is a constant which was chosen arbitrarily. This definition of the bracket operation satisfies translational invariance: for example, <(?o + ^oK#o + ^o)(*i+^iW'i+^i)> = <WoWi>= 1 because the terms multiplying the v's are all 0. Note also that the anticommutation rules mean that, for example, (WoW'i^1(3.33)
(In analogy to the scalar case, one requires W*o =  Wo> n o t W o = 1  Wo) One can now define the Feynman path integral on a lattice for the complete gauge theory including the Dirac fields. For example, the currentcurrent propagator on the lattice is
for any integrable function/ and any constant J„. It is this translational invariance that makes the Feynman path integral provide a realization of Schwinger's action principle (see, e.g., Ref. 12 for further discussion). Analogously one needs to define an integration over Fermi fields with the same translational invariance. Stated abstractly, one wants to define a bracket operation (' • •) defined on functions of purely anticommuting Fermi fields tpn with the property (A4>n+il„,yr,+V„))=(f(ipn,4>„)) , (3.31)
^ r v f n n f«m„)(?;„rp^wKwA),
(3.34) where A is the full action of Eq. (3.12) and
4 , = ( I I I I f dBmv)(eA) .
(3.35)
where rjn and rjn are anticommuting cnumbers (these have been introduced by Schwinger). The bracket operation should produce a number for every function/; it should also be a linear operation. Thus for a finite lattice it is sufficient to specify the bracket (• • •) for all monomials in the tyn and tfn. Because of the anticommutation rules, 4>„2 and Jp„2 vanish (more correctly, ty„a2 and lfna2 vanish where ip„a and $„n are any component of ip„ and ipn), therefore, there are only a finite number of possible monomials. It is now easy to see that
This formulation of the path integral is different from the formulation discussed in Sec. II. However, one can easily derive a lattice form of the path integrals of Sec. II fromthe present expression. The procedure is to expand Eq. (3.34) in powers of K, where K is the coefficient of the nearestneighbor coupling terms 4i„yjJii>ni.j;e'B"i' etc. This nearestneighbor coupling term can be represented diagrammatically by a line from the site n to the site n + p.. The expansion is best described by studying an example of a term from the expansion of the numerator of Eq. (3.34),which will now be discussed. An example of a term in the expansion is represented diagrammatically in Fig. 9. The expression for this term is
327
C O N F I N E M E N T O F QUARKS dBm\ K\l(;ny^lJmyvipJJ00y0tl>10e,B°o,°yioy^ue>aio.i ? u y 0 « 01 e ,fl oi.oft, iy ^ 00 e' a oo.i < ^ 0 > =D
(3.36)
10
2453
(OTJC
(D for diagram), where the four lattice sites involved are (w0, nx)  (0, 0), (1, 0), (0, 1), and (1,1); the values of ru and n3 are constant and have been suppressed in the notation of Eq. (3.36). The action A0 omits the K term, and is A>=
•r^' + 2?Ez 2g
2
(3.37)
The calculation of D has two parts, one being the integration over all gauge fields B m(i , the other being the calculation of the i/i, $ bracket. These are independent calculations, i.e., D factors into D^DB. The quantity DB is JOB = ( I I I T \ ,dBA
ex
p[ i ( 5 °o,o + Bio.i B oi.o£ 00 , 1 )+ 2 P £
e"»iiv\
(3.38)
This is an example of a guagefield average of the exponential of a line integral over a closed loop, the loop being the loop of Fig. 9. The tji bracket calculation can be factorized further into separate bracket calculations for each lattice site, since A0 contains no terms involving I/J or $ and coupling different lattice sites. Consider only the four lattice sites on the loop, for simplicity. By moving the iji's around some (using the anticommuting rule) the bracket becomes (? 0 ^o*io«"* l ^ w ?io>'i#i l e"* u * u ?uyM*iiWuyo*oie"*"* m ?oiy l *ooe^ , ^^oo>'»*oo> •
To make a product of all possible ip's and $'s means one must have products of all possible i ^ ' s and $oo's> all possible ^ 01 's and ^ 0 1 's, etc. In summary the complete bracket may be written as a product of four separate brackets. Define i)J=<* l o«"'*»«' l ^io> .
(3.39) (3.40)
Both D\ and Dtll are matrices in spin space due to the spinor indices implied for ip10) ^10, <plv and $ u . The full bracket is simply Dt =X 4 TptDjy 1 D f t l y 0 Djy 1 i) # 1 , r o ) . (3.41) The matrices D\ and Dn are easily determined. For example, D\ explicitly is £<«a = (Vioaexp (c £ 3W* l0 y) ? w } • 0.42)
if the nearestneighbor couplings combine to form closed loops (the lattice site at the endpoint of an open line would have an extra tpn or $„ so the bracket at n would give 0). The bracket calculation for a closed loop gives a trace involving K times a y matrix for each line in the loop and D's for each lattice site in" the loop (except the points n and 0 where there are currents). The average over gauge fields involves an exponential of a sum of .B„M's around each loop. There can be any number of loops. IV. STRONGCOUPLING APPROXIMATION The gauge field average lip) which determines whether quarks are bound was defined on a lattice in Sec. HI (Eq. 3.13). There are two limits in which this average can be calculated. The most interesting limit is the strongcoupling limit g~«>. This is the limit which exhibits quark binding. A strongcoupling expansion will be derived in this section.
(0,1) (1,1)
The exponential can be expanded in powers of c; assuming the spinors have four components only the c 3 term can produce a product of all four times all four ? l 0 's; the result is £*aa=5aa(c)3 (3.43) (the minus sign comes from the convention that the bracket is positive when ip10S appears to the left of Vioa) A similar calculation gives D*„=c%. (3.44) The results of this example are easily generalized. A term of general order K' is nonzero only
(0,0)
(1,0)
FIG. 9. Elementary square on the lattice.
328
2454 K E N N E T H G. W I L S O N H Expanding in p o w e r s of 1/g2, t e r m is the z e r o t h  o r d e r H)
The s t r o n g  c o u p l i n g expansion will be the b a s i s for a r e f o r m u l a t i o n of the gaugefield theory as a s t r i n g model. This will also be explained in this section. Consider specifically the n u m e r a t o r of Eq. (3.13), to be denoted IN(P):
mp)= ( n n X d B m )
exp i
[ ' ?{±}B""] •
(4.2)
\ m
v
JTT
/
x exp ^ X (±)£„„ + 2J2 S
eif 1
" "'] •
(4.1)\
This t e r m v a n i s h e s , since for any BnjJ which a p p e a r s in J^p, t h e r e is an integral jl^dB^ x exp(± z'B„p) which is z e r o . Thus one must seek h i g h e r  o r d e r t e r m s in g'2 which cancel the Bnjl in the line i n t e g r a l . The f i r s t  o r d e r t e r m is
W ) = ^ ( n n X /
B
 )
^exp[* 2
Ct)SnM+«/,„] .
(4.3)
The q u a n t i t y / ( i r o is itself a line integral of the gauge field; it is the line i n t e g r a l around a s q u a r e o r i g i n a t ing at the lattice s i t e I of s i z e a (unit s q u a r e ) . The integral for I(p{P) will vanish unless it is possible to find a unit s q u a r e such that flT0 cancels completely the line integral 2 P ( ± ) B „  r This is possible only if the path P is itself a unit s q u a r e . Otherwise the f i r s t  o r d e r t e r m vanishes and one must study the t e r m s of o r d e r g~* or higher. The t e r m of o r d e r g'2k has the form
/( , =
^ n(i£y(5PJ>"^
(4 4)
'
The only nonzero t e r m s in this sum a r e those for which ^(±)BnM+/(i,iOi
+
+/lft7rftO(! = 0 .
(4.5)
[See Eq. (3.8) for the definition of / l 7 r o in t e r m s of Bnjl.\ This equation can be understood g e o m e t r i c a l l y . E a c h / J l r o c o r r e s p o n d s to a s q u a r e of s i z e a on the l a t t i c e . F o r this sum to vanish the s e t of s q u a r e s defined b y / , ^ ^  • f,k ^ ^ must combine to make a surface with boundary P. (To be p r e c i s e , e a c h / ( T r o c o r r e s p o n d s to a line i n t e g r a l around a s q u a r e , and when these s q u a r e s a r e joined to m a k e a s u r f a c e the line i n t e g r a l s must cancel along all i n t e r n a l lines of the surface. The line i n t e g r a l s along the boundary P of the surface must run in the opposite d i r e c t i o n to the original path P.) See Fig. 10. F o r a given path P the lowest nonzero o r d e r in IN(P) is d e t e r m i n e d by the m i n i m a l a r e a A e n closed by P, the a r e a A being the a r e a of any s u r face built of unit s q u a r e s on the lattice with boundary P . Then JN{P)~(g2)'A/a , a p a r t from a n u m e r i c a l factor. This is the r e s u l t p r o m i s e d in Sec. II: The gaugefield a v e r a g e IN{P) behaves a s e x p [  j 4 ( l n ^ 2 ) / a 2 ] , i.e., exponentially in the a r e a enclosed by P. Hence, according to the a r g u ments of Sec. II, quark paths will not s e p a r a t e
m a c r o s c o p i c a l l y , and t h e r e will be no quarks among the f i n a l  s t a t e p a r t i c l e s . Consider h i g h e r  o r d e r t e r m s in the expansion of IN{P) for given P. T h e r e a r e many such t e r m s because t h e r e a r e many s u r f a c e s with boundary P. In p a r t i c u l a r , t h e r e a r e many ways to combine s u b s e t s off's to add to z e r o so such s u b s e t s can be added to any minimal s u m o f / ' s which f o r m s a surface with boundary P. The s i m p l e s t example of a set o f / ' s which add to z e r o a r e the set o f / ' s corresponding to the six faces of a unit cube. Written out, this gives
^ ~ _/nfi V ~Jn+ ir ,11 V + / nil 71 ~Jn+ (j , v n +J n^il ~ J i t t i / , ^ >
(4.6) which is easily checked using Eq. (3.8). [Equat ion
FIG. 10. Filling of enclosed area of path P by elementary squares.
329
CONFINEMENT OF QUARKS 2455
10
(4.6) is the lattice analog of the equation Let A(P) be the minimal a r e a as defined enclosed by P. Since one can place a unit anywhere on t h i s minimal a r e a , it m e a n s t h e r e a r e roughly A(P)/a2 m o r e t e r m s of
gl2(g2)A(P)/a2
above cube that order
^
t h e
e x p a n g i o n
Qf 7
^
p )
t h a n
t h e r e a r e t e r m s of o r d e r (g2)'A(p)/a2. [One can place unit cubes anywhere in space, not j u s t on the m i n i m a l s u r f a c e ; but when one divides IN{P) by Z all disconnected t e r m s cancel, a s u s u a l . ] This s u g g e s t s that the 1/g2 expansion is not very useful in the limit A(P)*>, which is the limit of i n t e r e s t for q u a r k binding. However, experience with r e l a t e d p r o b l e m s s u g g e s t s that IN{P) is not the a p p r o p r i a t e quantity to expand; instead one should t r y writing I(P) =
=
the q u a r k loops. The sum of the g'2 expansion is a sum over all such s u r f a c e s . This is p r e c i s e l y the s t r u c t u r e appearing in s t r i n g m o d e l s : c o m bined s u m s over quark loops and interpolating s u r f a c e s . However, the loops and s u r f a c e s of the gaugefield theory a r e defined on a lattice w h e r e as the loops and s u r f a c e s of the s t r i n g models a r e defined on a continuum. It may not be easy to d e r i v e quantitative r e l a t i o n s between the two types of s u r f a c e s . V. WEAKCOUPLING APPROXIMATION The weakcoupling approximation will be d i s cussed briefly, leaving many questions open. Only the p u r e gauge field will be d i s c u s s e d . Consider again the e x p r e s s i o n
Z1IN(P)
(g2)MP)ec(p,e*)
> (4i7)
^)=(nn/* ds «)
\ m
v
J
TT
/
and expand c(P,g2) in p o w e r s of g"2 instead. One would expect c(P,g2) to be dominated by a t e r m p r o p o r t i o n a l to A(P), say c(P,g2)=A(P)f(g2) + 0(P) (4.8)
xexp \i £
L P
(±)B„, + A
E
e""" "1 •
J
•Si'nil V
(5.1) Suppose the integration v a r i a b l e s w e r e /„ M „ r a t h e r than Bmv. F o r s m a l l g, only s m a l l values of/„ [1 „ would be i m p o r t a n t in the integral, in o r d e r that B.ee'fmv be near its m a x i m u m value 1. One would then expand:
(where P is the length of the path P). The c r u c i a l question i s the n a t u r e of the s e r i e s for f(g'2). P a s t e x p e r i e n c e with s i m i l a r types of expansions (namely, the h i g h  t e m p e r a t u r e expansions of s t a t i s t i c a l m e c h a n i c s : s e e , e.g., Ref. 13) s u g g e s t s that f(g~2) will have a convergent expansion at l e a s t for g'2 l e s s than a c r i t i c a l value gc ~2. Howe v e r , no calculations have been done in the gaugefield theory for f{g~2) a s yet. C o n s i d e r the complete expansion of I{P). Each nonzero t e r m in ±he expansion c o r r e s p o n d s to a surface with p e r i m e t e r P. The complete expansion c o r r e s p o n d s to a sum o v e r all possible s u r faces with given p e r i m e t e r P. "All p o s s i b l e " s u r faces include s u r f a c e s which i n t e r s e c t t h e m s e l v e s (to take into account t e r m s w h e r e a givenf nflv a p p e a r s s e v e r a l t i m e s in the s u m / , 0 + •• • + f'k"k ai)' T h e r e is a weight factor for each s u r face, aside from the power of g'2 d e t e r m i n e d by the a r e a of the s u r f a c e . F o r a simple surface, the weight factor is 1; the weight i s m o r e c o m p l i cated for s e l f  i n t e r s e c t i n g s u r f a c e s . Thus, the strongcoupling expansion for the c u r r e n t  c u r r e n t p r o p a g a t o r h a s the s a m e general s t r u c t u r e a s in s t r i n g models of h a d r o n s . One is actually dealing h e r e with a double expansion. An expansion in the coefficient K (appearing in the D i r a c field action) was needed to define quark loops on the lattice; the sum of the K expansion i s a s u m over all possible quark loops. The g'2 expansion is needed to define s u r f a c e s filling in
With this approximation one could extend the limits of integration o n / , , ^ from ±it to ±°°, with negligible e r r o r ; one would then have a s e t of G a u s s i a n i n t e g r a l s to evaluate. In p r a c t i c e the integration v a r i a b l e s a r e the Bmv, not t h e / n  J „ . However, one can make a change of v a r i a b l e from the Bmv to t h e / n ( J „ . It is not possible to eliminate all the BmV by this t r a n s formation, and not all the v a r i a b l e s / , , ^ a r e independent. N e v e r t h e l e s s , the t r a n s f o r m a t i o n is sufficient to make IX(P) calculable for s m a l l g2. To make the change of v a r i a b l e s p r e c i s e , cons i d e r a s y s t e m of finite s i z e ( 1 « nt « N) with periodic boundary conditions. Then one can change v a r i a b l e s from the BmV to a subset of the fm v plus s o m e gauge t r a n s f o r m a t i o n v a r i a b l e s <p„, plus four e x t r a v a r i a b l e s £„, a s follows:
(i) F o r >I0TN, «!, n2, n3 a r b i t r a r y , Bn)1 (JJ. = 1, 2, 3)
is r e p l a c e d b y / „ M 0 . For B„0, one w r i t e s Bna = 4>n+S4>n (5.3)
and r e p l a c e s B„0 by </>„. This is the e s s e n c e of the t r a n s f o r m a t i o n from Bm t o / n / J 0 for u. * 0 and from Bnn to e>„. To complete the t r a n s f o r m a t i o n one must d i s c u s s the surface n0 = N.
vacuum expectation values decrease rapidly at separations of only a few lattice sites (there is a factor g~2 or . where 5 n 3 = 0„+3. nt±N. BnjJ (H5t3)is replaced by/ n(j3 . If the propagator behaves as g~2" for distances x = na.(P) well defined and equal to I N{P). also there is no convergence factor for the £ integration. There should be a transition between these two phases which would occur at a critical value ge for any g and any small value of K. using the same change of variables as for IK{P). n2*N. This integral can be computed by explicit Gaussian integration methods rather more easily than the/ n(1 „ integrations for IN(P). one is interested in practice in values of g and K such that the correlation length i is much larger than the lattice spacing a.e. This is one possible phase transition. To make If.. then the corresponding mass is 2{Vag)/a. a term ~ip{"ZjnBnll)2 and (b) add a quadratic form in t h e / ' s to compensate for the result of the <p integration of the gaugefixing term. (5. PHASE TRANSITIONS In the strongcoupling limit (g*<*>.^ variables using Eq. a strongcoupling phase for large g which binds quarks. = »! =nz = N. (iv) For «(. This is larger than the cutoff momentum ir/a if g is large. [The usual rule is that if a propagator falls as e~*'% for x large then the lowest mass intermediate state contributing to the propagator has mass 1/1.. Bnfl (ix ±2) is replaced by/„^. as will be argued below. Suppose. one starts with i ^ = ( T O £ «) p (* 5 » . 14 It has already been argued that there are two distinct phases for the gauge field. since the action was defined on a Euclidean metric.. it is this transition which was discussed in Sec.6). and rc3 arbitrary. one finds (£„ is present only if mv =N): + (linear combination of/„. II. Thus one seeks special values gc and Kc for g and K at which there is a phase transition. neither the <(>„ nor JM appear in these expressions. but with infinite limits of integration. The result is different in this case due to the a term which couples the 0's to t h e / ' s . it is Euclidean invariance that is missing. WILSON 10 (ii) For n0 = N. but since the net result is still that IN(P)=I'N{P) is a Gaussian integration in the B's. in order that the corresponding mass is much less than the cutoff. for n0±N) can be expressed in terms of the independent/.4) (iii) For «„ = »! =JV. This corresponds to the existence of masses much larger than the cutoff. B„u (/i #1) is replaced by/ n/J1 . One knows from statistical mechanics that large correlation lengths are associated with secondorder phase transitions (critical points). (5. T h e / integrations are nontrivial because of the constraint (4.. What one wants to accomplish is to reduce the lattice theory for small g to something like a conventional free gaugefield theory. which is reasonable if K is small. The arguments given neglected quark vacuum loops.)2 gaugefixing term. and with a gaugefixing term included. (v) For n0 = nl=n2 = n3=N one writes Bnfl = £p with f being the new variables. and n2 and n3 arbitrary. with B„2 = <p„+z(t>„. £„ 3 by <£„. VI. K~0) the gauge theory is far from being Lorentzinvariant. IK(P) can also be reduced to an integral over a subset of the/ n(1 „.0„.e. When an arbitrary Bmv is expressed in terms of the new independent variables. Bnl is replaced by <Pn. and a weakcoupling phase for small g which does not bind quarks.„) .5) i. It is easily verified that the integrand of IN(P) involves only t h e / ' s : It is independent of both the <p's and £„ (the latter does not appear because any closed path P has as many Bnv terms as +Bm terms on the sublattice nv=N). II. The variables /B(I1/ which are not integration variables (for example.330 2456 K E N N E T H G.] Thus.. This means restoring the Bmv as the integration variables.» • I iF £ " ) • " K or both for each unit lattice spacing of separation). for example. Hence the <pn a n ^ £„ integrations can be computed trivially. the result will presumably be similar to the conventional freefield calculation reported in Sec. The author has not carried through this calculation. i. In the strongcoupling limit. and n3±N.f ? [ ? "» . In addition. fnjl v with y. Bn2 is replaced by 0 „ . But. the <p variables define a gauge transformation and £„ represents a translation of some of the B's. one must (a) put in a convergence factor for the £ integral. this is probably a firstorder tran ds ex (±)B {B B } _ u 2 where the a term is a lattice version of a (V^A. One also sets <> = 0 /„ M for this value of n.2. with Bnl = <j>n+:<p„ . * 0 and y * 0.6). (4. More precisely.
S» ' (6. SU(3)xSU(3) is an exact symmetry rather than a spontaneously broken symmetry. H this does not work one is free to introduce additional terms into the quark field action in hopes of forcing a spontaneous breaking of SU(3)xSU(3). M = 0 for A = 0. (2) series expansions. The renormalizationgroup approach is potentially the most powerful and accurate method for studying lattice theories near a critical point. namely M = Z0"1 £ s^'^'o 2 = V J0=il e(2*KHHl)s0 .331 C O N F I N E M E N T O F QUARKS 2457 10 sition rather than second order. K~0). 16 for one of the best seriesexpansion formalisms.4) with d being the dimensionality (3 usually) and _ (6. One then uses Padeapproximant techniques to look for singularities in either g or K that would be a s  sociated with a mass approaching 0. Then there will be a phase transition at a critical value KQ for K where one changes from exact SU(3)xSU(3) to spontaneously broken SU(3)xSU(3). and (3) the renormalizationgroup approach.5) The result is M = tanh(2dKM+ft) . In the remainder of this section these methods will be discussed briefly. 13 for a general review. but at present the renormalizationgroup techniques are too limited in scope to be applicable to the present problem. in this case the theory might be a realistic model of broken SU(3)xSU(3) for K slightly greater than Ka (with g large enough to maintain quark binding).0 . If this transition is a secondorder transition then there will be a large correlation length for K near Ka. to high order in g'2 and K. they apply mainly to propagators. Various methods have been developed by statistical mechanicians to deal with this problem. stability considerations show that one must choose a solution with M * 0 when h = 0. and to give rough estimates of the behavior near the transition. but one could hope to generate maybe 6 or 7 orders with some practice. that SU(3) xSU(3) can be broken by increasing K. . Meanfield techniques 15 are the simplest and crudest methods for studying a critical point. For 2dK>l the solution is multiplevalued. Examples of meanfield calculations will be given later. It is especially difficult to solve the lattice theory near a critical point with a large cor relation length. Then AfZ»(s 0 exp ( * ] £ ] £ s .6) If 2dK< 1 this equation has a unique solution for M as a function of h. In simple statisticalmechanical problems one can generate 12 terms or so in analogous expansions. whether it is first or second order. Return to meanfield ideas. See Refs. see Ref. in particular. Apart from special limits (g"*> and K . and one must have a clear idea of what one is trying to learn before attempting such calculations. (6. In summary. As a result. being very awkward to perform on threeand fourpoint functions. 6 and 17. and 2 = ( e x p ( A ' E E s " V + A j s „ J . or g small) it is very difficult to solve the lattice theory. (6. so one must hope that by increasing K one can change the exact SU(3)xSU(3) into broken SU(3) XSU(3). The expansion for the lattice theory of this paper is more complicated. A little thought shows that in the strongcoupling limit (g°°. Series expansions require considerably more effort than meanfield calculations. let the interaction be TT=*ZZ H S S n n+Z+kT. the formula for M simplifies to a sum over s0 only. None of the results of a meanfield calculation are entirely trustworthy.2) where ( • • •) means a sum over all configurations of all spins.1) where K is related to the spinspin coupling and h is proportional to the external field. one assumes that the spins sj coupled to s 0 can be replaced by their average value M. the transition of real interest is a transition in K (or some other parameter introduced into the quark action) rather than g. Varying g does not change this situation. while the quark fields would carry SU(3)xSU(3) quantum numbers as well as gaugegroup quantum numbers. An example of a series expansion would be the expansion of the currentcurrent propagator for small momentum (momentum « 1 / a ) in powers of g'2 and K. 15 The prototype meanfield calculation is a calculation of the magnetization as a function of the external field for an Ising ferromagnet. There are essentially three approaches to consider: (1) meanfield techniques. s ^ j + A E s « ) ) ' (6. Then the gauge fields would all be SU(3)xSU(3) singlets. See Ref. for simplicity. Suppose. Suppose one wishes to construct a model of strong interactions using the lattice theory of this paper with the gauge group separate from ordinary SU (3)xSU(3) symmetry. Let s„ be the spin at site n with values ±1 only.3) In the meanfield approximation. (6. invariably they are the methods one uses first in studying a new situation. They are used to determine if there is a phase transition.
L. 2 J. Phys. Weisskopf. 205 (1973). suri. proceedings of the International Conference on Symmetries and Quark Models. B. Yan. In the gaugefield theory e'B°t> couples to a product of t h r e e other exponentials. Prog. P. 13 M. 3501 (1974). Sec. Y. Phys. Susskind. New York. F.g. Jaffe. H. However. Rev. 12£. 457 (1970). e.. Schwinger. Phys.332 2458 KENNETH G. 50B. E. The Method of Second Quantization (Academic. Rev. Amati and M. 255 (1974). which is a good approximation if d is l a r g e . 75 (1974). thesis. Theor. L. and L. R. Rev. 615 (1967). Drouffe. A. Lett. Nuovo Cimento 69A. "See. T> 1_. Phys.o r d e r t r a n s i t i o n at the value of g for which M changes from z e r o to being n o n z e r o . P . and V. and Itzykson. Rev. F i s h e r . M. P. and stability c o n s i d e r a tions show again that the A / * 0 solutions a r e p r e ferred. Nucl. and m e m b e r s of a s e m i n a r at O r s a y .M . Nambu. Thus t h e r e i s a f i r s t . 10 See. Phys. G r o s s . Wilson. 2386 (1973). Johnson. Norton. Johnson. Testa. S. Susskind. 1 9 A c l e a r review of q u a r k confinement in the lattice theory is given in Ref. 9. J .L. Coleman and E. 2 1 . Rep. 9C. 5 A. G. Sakita. J. Phys. 2425 (1962). 227 (1974). D. 3338 (1973). Wilson and J. Lowenstein and J. K. 397 (1962). Nucl. B64. Ann. Rev. 419 (1972). WILSON 10 In this approximation one has actually r e p l a c e d J^^s. the gaugefield c a s e never has a solution for h = 0 with M s m a l l but nonzero. Phys. 1 (1973). D. R. Phys. Phys. Phys. R. The r e s u l t of this i s that M=f f{dD g' M3+h (6. e.g. a s a meanfield approximation one r e p l a c e s this p r o d uct by M3. 9 R. 1966). S. S_. M.8) ACKNOWLEDGMENTS The author h a s benefited from c o n v e r s a t i o n s with many people in developing and e s p e c i a l l y in understanding the lattice gauge theory. Chodos. Phys. Rebbi. A. B. Rep. Rev. 1 C o" dBoveiB°» xexp ( \ „„1 2g2 M3 + h) (eiBon+eiB°u) (6. Cornell University. Cornwall and R. 172 (1971). Konisi. Feynman. D 10. New York. J. Phys. A much m o r e thorough d i s c u s s i o n of the m e a n field approximation h a s been given by Balian. (6. Phys. Susskind. B (to be published). X. S K. T .. 128.7) w h e r e d is the s p a c e . Casher. Fisher. S. (N. Lett. Olesen. 716 (1973). Weinberg. 4 A. Phys. 'See.fl e oji. Berezin. and M can be defined to be the expectation value of . 14 For the relation of relativistic field theory to critical 8 7 . + h) (e'a0M+«'%)] (6. L. edited by Ramesh Chand (Gordon and Breach. 20.g.. P. Rev. An analogous meanfield calculation can be p e r formed for the lattice gauge theory. In the magnetic c a s e . Thorn. The r e s u l t of this is that M = Z.) 68. A. Rev. Wayne State Univ. Kogut. 1969. If g is l a r g e the solution to this equation is unique and *Work supported in part by the National Science Foundation. one finds that the s p o n taneous magnetization M goes to z e r o for 2dk — 1 [from Eq. 440 (1950). Phys. 48. J. Ph. Susskind. 1970).Y. If g i s s m a l l then t h e r e a r e s o l u tions with M * 0 for h = 0 . E. Mandelstam. Bjorken. In this c a s e a simple e x t e r n a l field t e r m has the form h~Y^ {eiB"tL+e~iB"») (this is to be added to the gaugefield action). Goddard. 3 J.9) w h e r e / i s a r a t i o of B e s s e l ' s functions. 52ff. Phys. 109 (1972). Lett. ibid. Lett. 732 (1974). 1969 (unpublished). 1888 (1973). 80. The question is whether M i s z e r o in the limit h~ 0. G. Abers and B. W. P e r s o n s I a m indebted to include J . Goldstone. Phys. E. Kogut and L. C. and C. Another formulation of the connection between strongly coupled gauge t h e o r i e s and s t r i n g models is given in Ref. 3471 (1974). Prog. B56. 31. Jackiw and K.by 2dM. Swieca. in' Symmetries and Quark Models. 2008 (1972). 30. Rev. 48E. Thorn. A nonzero value of M in the l i m i t h — 0 m e a n s one has spontaneous b r e a k i n g of the gaugefield s y m m e t r y . G. 3_0. D 8_. Gervais and B. and Z„= J" expU^f^M J0=  expl 3 M = 0 for h=0. 125.. Kogut. 12 K. E. C. Nucl.6)]. Rep. ibid. Phys. D 9. D 6. G.t i m e dimensionality. . Phys. e. 't Hooft. J. pp. Kogut. J. 1 8 A Hamiltonian f o r m u l a tion of the lattice gauge theory h a s been given by Kogut and Susskind. So for s m a l l g the theory shows s p o n taneous b r e a k i n g .D. 792 (1973). F. F e y n man. Lee. T h i s i s generally t r u e of meanfield theories.
Brout. J. Tassie. B61. Phys. Sec. 6. Rev. Cornell Report No. Itzykson. Phase Transitions (Benjamin. Rev. 176. Phys. 21 H. Nielsen and P. 1974). 46B. Olesen. Cargese (1973) Lecture Notes. 739 (1968). Phys. 1965).333 C O N F I N E M E N T O F QUARKS 19 10 2459 olnts see Ref. New York. CLNS271 (to be published in the proceedings of the conference on YangMills Fields. L. 1SD jasnow and M. . Rev. J. Wilson. and references cited therein. D (to be published). 17 K G. 18 R Balian. 20 K. Nucl. Phys. in preparation. p. 45 (1973). B. D (to be published). Kogut and L. M. 397 (1973). Phys. Lett. Marseille. X. Wilson. Wortis. J. Susskind. Drouffe. and C. " F o r a review see R. 8.
i. S[A]<~. Due to these random fields. Indeed. occurring with a gauge field. it is a good and simple model for trying our program on. (In fact. the compactness of quantum electrodynamics seems to be an attractive hypothesis and our results may have physical applications. POLYAKOV Landau Institute for Theoretical Physics. . This is precisely the phenomena we are going to investigate. Now assume that the fields A^ are such as if they were produced by certain "particles" in the fourdimensional euclidean space. Of course. though the weight with which non82 (where E is certain constant) their influence on the cor relation is large if the classical field A is long ranged. for very large R. Moscow. might provide a natural explanation for the confinement of quarks. It is the purpose of the present paper to work out a formalism which permits. This means averaging over all possible fields A„ with the weight equal to: exp{S(/l)} = exp F trivial minima enters the averaging is small being proportional to exp {S(A)} = exp(E/g2) (3) ^ S p J>M2„ d 4 x I 4.e. The above discussion was based on the crucial assumption that there exists pseudoparticle solutions of the gauge field equations. (2) Usually one takes into account only the trivial minima of 5. The existence of such solutions can lead to afiniteconelation length which stops infrared catastrophe.< 1. then the leading role in the averaging will be played by the fields close to that defined by the equation: rj7T=0. It was pointed out by different authors [1] several years ago that the infrared phenomena. The purpose of this consideration is two fold. It will be proved in the second paper of this series that such solutions indeed exist for every compact nonabelian gauge group. >4M = 0.334 Volume 59B. at least partly.' HV^'bnAv'dvAn + j • (1) [An>Av\ 2 Assume that the charge g. First. to take into account the infrared effects in gaugefield interactions. Hence. The average density of pseudoparticles in our system is very small. being proportional to exp(—E/g2). For example we shall prove the existence of a certain critical charge in QED. the "energy" E depends on the number of the above mentioned pseudoparticles. In the present paper we deal only with theories with a compact but abelian gauge group. For the correlation function with the distance R the parameter of the perturbation expansion isg2 log R/a where a is the inverse cutoff. In this first paper we confine ourselves to the problem of realizing the above program in the case of compact but abelian gauge fields. their existence creates long range random fields in our system. Second. the contribution to the correlation will be shown to be proportional to exp{—Zf/g2}/?4). In other words there exist the "oneparticle" minima of 5". At the same time there exist no methods for analyzing the interaction of gauge fields in the deep infrared region.M. USSR Received 19 August 1975 It is shown that infrared phenomena in the gauge theories are guided by certain classical solutions of the YangMills equations. and developes the perturbation theory as a small deviation from this. perturbation theory is not applicable and another/f might become essential. number 1 PHYSICS LETTERS 13 October 1975 COMPACT GAUGE FIELDS AND THE INFRARED CATASTROPHE A. In this case the problems of correlation length and charge confinement are completely solved. the correlation length becomes finite. Our main idea is that the system of gauge fields acquires a finite correlation length through the following phenomenon. Imagine that we are calculating a certain correlation function in the euclidean formulation of the gauge theory. However. the "twoparticle" and so on.
number 1 PHYSICS LETTERS 13 October 1975 The definition of the theory is as follows. For the evaluation of (9) let us substitute A^ = Afl + aM. The sum in (10) goes over all possible configurations of pseudoparticles. Hence. Now. (14) was derived for arbitrary planar contour.v IV' (3) ^al f. due to the periodicity of the exponent. If we introduce the field: a 2 ea$y F@y in which. The action is given by: S(A)=E/g2 (8) 5= TJ f(Fx x. Generally. f(x) * x0 1 4g2 (the value of the constant e depends on the lattice type and is not essential for us). Eq. there exist the Debye correlation length and the corresponding photon mass m equal to: m 2 = exp{e/g 2 } (in the units of the inverse lattice length). In this case there exists quasiparticle solutions of Maxwell equations which simply coincide with the Dirac monopole solution. It will be convenient for us to analyze first the three dimensional theory. According to 83 . due to the periodicity. let us use the formula: exp{i#^ dx } = e x p ( i / F a daa] 01) ^W=C + 2 »^J ( S '%) (5) where 6 ( S ) (x) is the surface 6function. (4) Gauge theories on the lattice have been considered earlier by Wilson [2] and the present author (unpublished).* a*b \ ~ b\ x x a where aM is lattice vector. Since the integral over aM is gaussian we get: 2exp{SC4)}exp{i!M dxj ^C) =F (C) ° SexpMU)} 00) (Here F0 is the contribution of A = 0). The result is two fold.z. the action should have the form: If {qa} are integers then the singularities in (7) are just of the permitted type. necessary in the definition of functional integrals.2775a3 Tf qa0(x3 x3a)8(xxla)8(x  x^). and the group is the circle and not the line. See also [3]. The problem is reduced now to the calculation of the free energy of the monopoles plasma with the "temperature" # 2 in the external field: (6) This problem was solved by using Debye method which is correct for sufficiently small g2. Secondly: W[C] = coitst(g2mA) (14) (13) then the general pseudoparticle solution will be given by: F _ vVfr~*A " 2 \XXa\ 3 (7) .n.fE^+. where A is the area of the contour C. in the continuous limit F^ may have the following singularities: (here Cis some large contour). This is equivalent to the hypothesis that: / ( * +2TT) = / ( * ) .Volume 59B. Let us introduce a lattice in the four dimensional space. only the first term from (7) should be substituted. Now let us analyze the correlation function introduced in [2] which is most convenient in the confinement problem: F(C) = ex P {H/(C)} = (exp{i!M M dx M }) (9) The hypothesis of the compactness of the gauge group means thatAXfl are the angular variables. The immediate consequence of the periodicity of f{x) is that the nearest neighbours Ax+ax and Ax_M can be different by 2TTN (where N is integer) without producing large action. The second term in (5) will not contribute to the action. First.
Wilezek. In the case of the four dimensional QED it can be shown that the only classical solutions with finite action are closed rings. Gross and F. Phys. D10 (1974) 3376. Phys. after small variation of the x 4 . Lett. Since it was proved in [2] that for large g2 the charge confinement exist there are some critical charge g% at which the phase transition occurs. Phys. This follows from the fact that singular points in this case should form lines. and B is some constant. our pseudoparticle will be outside the cube. Then it should be: #/V„ dffM„ = 2mf (15) where L is the length of the contour C. Consider the cube K with x 4 = 0. Rev. It is not clear now whether this critical charge is connected with the fine structure constant. 31 (1973) 494. D8 (1973) 3633. Rev. Since the closed rings produce only dipole forces their influence on the correlation are rather weak. D.336 Volume 59B. [3] R. 84 . Wilson. This result means the absence of the charge confinement for smallg 2 . The extension of the above ideas on the nonabelian theory will be presented in the other papers of this series. [2] K. and consider the pseudo particle solution with x = 0. Weinberg. Rev. We showed that in this case the correlation length remains infinite and that W[C\ = const exp(B/g2)L (16) References [1] S. Balian et al. assume that it is not so. Phys. number 1 PHYSICS ITERS 13 October 1975 Wilson [1] this result means "charge confinement" in the three dimensional QED with the compact gauge group. D10 (1974) 2445. But. To prove this.. Rev. and this contradicts (15).
It is shown that only planar diagrams with the quarks at the edges dominate. or infinite.Nuclear Physics B72 (1974) 4 6 1 . Sakita and Mandelstam for the dual string. Introduction The question we ask ourselves in this paper is how to construct a field theory of strong interactions in which quarks form inseparable bound states. First we have the singular infrared behaviour of massless gauge theories [1] that makes it impossible to describe their spectra of physical particles by means of a perturbation expansion with respect to the coupling constant. It is natural to take the symmetry corresponding to this quantum number to be a local gauge symmetry of some group SU(JV). If we consider the parameter T of the colour gauge group SU(A0 as a free parameter. a formal argument in terms of functional integrals has been given by Amati and Testa [3] that supports the conjecture that "coloured" states have infinite energy. The analogy with the string can be pursued one step further by writing the planar dia . but we do wish to point out some remarkable features of certain (gauge) field theories that make them an interesting candidate for such a theory.4 7 3 . energy might be required to create a physical state with nonzero "colour" quantum number. then an exV pansion of the amplitudes at N >• °° arranges the Feynman diagrams into sets which have exactly the topology of the quantized dual string with quarks at its ends. Geneva Received 21 December 1973 Abstract: A gauge theory with colour gauge group UOV) and quarks having a colour index runnin from one to N is considered in the limit N'>• °°. such that the number 1 IN corresponds to the dual coupling constant. In that case. A mathematical framework is proposed to link these concepts of planar diagrams with the functional integrals of Gervais. It is not inconceivable that in an infrared unstable theory long range forces will accumulate to form infinite potential wells for single quarks in hadrons. We do not claim to have a satisfactory solution to that problem. 't HOOFT CERN. In this paper we put the emphasis on an interesting coincidence. NorthHolland Publishing Company A PLANAR DIAGRAM THEORY FOR STRONG INTERACTIONS G.g N fixed. The HanNambu quark theory [2] gives a qualitative picture of such forces between quarks: a very high. For hadrons N is probably equal to three. 1. the topological structure of the perturbation series in 1/yVis identical to that of the dual models.
Planar diagram theory grams in the light cone reference frame. U (N) gauge theory In order to show that the set of planar diagrams may play a leading role if certain physical parameters have certain values.3 A.'+g[A . But for the time being it is there and we must keep it in mind when we finally interpret the results of our calculations. but it does simplify the arguments in sect.A ]J.*''(3 4>. 6.4) For sake of simplicity we do not make the restriction that the trace of the gauge field. (2.. or when we switch on weak and electromagnetic interactions through the Higgs mechanism.' (x) = A*. we write down a Hamiltonian that generates all planar diagrams.form three representations of the group U(N).1) and the Lagrangian is £ = \G where G . and so we will have a photon corresponding to the Abelian subgroup U(l) of U(7V).. 2. t Hooft. (2.5) We do not know whether this assumption is really essential for the theory. n and X. (2.' =3 A. There is an antiHermitian gauge (vector) field A. We add to the Lagrangrian  9 A. The Feynman rules [4.2) The index a runs from one to three ql=p\ q2 = n. A'.462 G.' . but that he cannot distinguish the different colour components (see also sect. .and X. 3) *.' + g [A .. in a Hilbert space of a fixed number of quarks.1 3 0.iq^ijD + m. In sect. . Of course we could dispose of it. Let us take the Feynman gauge. .. should vanish. The quarks Pp n. Let us assume that an observer can distinguish between p.1' d A. 0 ] A (2. either by replacing U(JV) by SU(7V)...N. The quarks are inseparable if and only if the spectrum of this Hamihonian becomes discrete in the presence of the interactions. 3.i= 1. we first formulate a possible gauge theory for strong interactions in which the parameters N and g have arbitrary values.. q3 = X.)qa. and coupling to baryon number.' G .i (x). 5] may be formulated as usual in any suitable gauge.
Feynman rules for V(N) gauge theory in Feynman gauge. ghost) ——. Planar diagram theory 463 (vector field) J A V J k2ie I (F.339 G. 1. 6 a afl S MI> 6. .P.) KH H + iykie i' ig{& (kq) +6 (pk) +6 (qp) } r 2 {25 an Pv 6. 't Hooft. m^ i J — (quark q. 0M 5 avJ } A * w> £7' ^ '«p„ ^ /* £7„ /r 'I 1 (Fermi statistics) Fig.
The extra minus sign in this propagator is a consequence of the antiHermiticity of the field ^4 (eq. . and that is when an indexline closes.'=N. an A • 'propagator to the left if/'</and a real propagator if/' =/. The ghost fields satisfy no Hermiticity condition and therefore their propagators have an additional arrow (fig. and a lower index by an outgoing arrow. with edges formed by the quark lines. We close the surface by also attaching little surfaces to the quark loops separately. because of the summation in (3. The quark propagators consist of a single line. for instance EP'V /' (3. but they continue. Planar diagram theory where 0 is the FeynmanDeWittFaddeevPopov ghost field. 2 we assumed that the observer is colourblind. We are now in the position that we can classify the diagrams with gauge invariant sources according to their power of g and their power of TV. Now the bilinear parts of the Lagrangian generate the propagators and the interaction parts the vertices.1). it is convenient to split the fields A^iinto complex fields for /' > / and real fields for i = /'. In fig. Note now that the number T does not enter in fig. the number JV will enter into expressions for the amplitudes. In order to keep track of the indices. This can be formulated more precisely: only gaugeinvariant quantities can be measured. 1 (this would not be the case if we V would try to remove the photon). Such an index loop gives rise to a factor Z) 5. Let there be given a connected diagram. 1).1)). of course. (2. But. the vector propagator stands for an A^l propagator to the right if / > / . and thus connect ingoing with outgoing arrows. i 3. 1. We get a big surface. As usual. The N > °° limit In sect.464 G. One can then denote an upper index by an incoming arrow. 2 that index lines never stop at a gauge invariant external source. The vertices always consist of Kronecker delta functions connecting upper and lower indices. amplitudes and Green functions are obtained by adding all possible (planar and nonplanar) diagrams with their appropriate combinatory factors. First we consider the twodimensional structure obtained by attaching little surfaces to each index loop. and which is in general multiply connected (contains "worm holes"). "Index loops" going through an external source also obtain a factor N. The propagator is then denoted by a double line. 't Hooft. A measuring apparatus can formally be represented by a c number source function/(x) which is coupled to a gauge invariant current.i) We observe from fig.
7) If the sources are coupled to quarks. where Vn is the number of npoint vertices.2) By drawing a dot at each end of each internal line. The diagram is associated with a factor r=g N'. g2N = g2 (fixed). (3. and V vertices. 2. (3. And so. Note. that the above arguments not only apply to gauge fields but also to theories with a global U(7V) symmetry containing fields with two U(A0 indices. who considers gauge fields on a dense lattice and also finds structures with the topology of a two . they are the planar diagrams with the quark line at the edges (fig.). P internal lines or propagators. Here F= L +1. r = (g2N?V3 + V4 N22HL. (3. we find that the number of dots is 2P= ^JnVn.5) where H counts the number of "holes" in the surface and is therefore always positive (a sphere has H = 0. and V= 2„ Vn.2) can be written as r = g2P2VNFL ( 3 4 ) (3. Let that surface have F faces.G. n and eq. but from the introduction. 3).6) Suppose we take the limit A^°o.3) Now we apply a wellknown theorem of Euler: FP + V=22H. It is interesting to compare our result with that of Wilson [6]. (3. The leading diagrams in this limit have H = 0 and L . it will be clear why we concentrate mainly on gauge fields.1. Planar diagram theory 465 Fig. then there must be at least one quark loop: L 3 s 1. g^O. etc. (3. a torus H = 1. however. where L is the number of quark loops and I the number of index loops. Gauge invariant source function. 't Hooft.
. Two diagrams of higher order in l/N: (a) obtain a factor l/N. One of the leading diagrams for the fourpoint function. .466 G. Here we see that the analogy with dual models goes even further. Planar diagram theory Fig. The dual topology of the set of planar diagrams has been noted before [7]. the expansion in powers of l/N corresponds to the expansion with respect to the dual coupling constant in Fig. It is not difficult to show that also Wilson's surfaces are associated with factors l/N2 and 1/TVfor each worm hole or fermion loop. dimensional surface. 3. respectively. 't Hooft. (b) obtain a factor l/N as compared with the lowestorder graphs of the previous figure. 4.
Such a theory is infrared unstable [1] which implies that infrared divergences accummulate instead of cancel. (or SU(3)) for which no preferred reference frame in the form of a Higgs field exists. (4. replacing the vertices by simple local 0 3 or 0 4 interactions. in spite of the fact that N is not very big. If we adopt the HanNambu picture of hadrons [2] then N is very likely to be three.2) (for sign conventions. We believe that a more careful study of this problem is necessary. 4.7) does not describe the spectrum and the Smatrix. in order to be able to do the integrations. This seems to give a reasonable order of magnitude for the dual coupling constant.a> we perform an integration over its time coordi . We immediately face two problems: (i) how to find a convenient parametrization scheme to indicate a point of the graph in the plane. But the l/N expansion may be a reasonable perturbation expansion. We assume that there is a local gauge group of the type U(3). [5]). Planar diagrams in the lightcone frame The theory implies that we have to sum all planar diagrams in order to get the leading contributions to the amplitudes. V2 **= — (x3 ±x°). We go over to a mixed momentum coordinate representation: at each vertex V. x =(x\x2). 5). For a moment we shall abandon the rather complicated Feynman rules of fig. Attempts to calculate certain large planar diagrams are known in the literature [7] but it seems to us that the choice of diagrams there is rather arbitrary. in terms of two parameters a and T. (ii) how to arrive at Gaussian integrands. we shall consider the slightly more general case of arbitrary masses. see ref. A simpleminded perturbation expansion with respect to g 0 in eq. and the replacement of a propagator by Gaussian expressions seems to be a bad approximation.G. 't Hooft. Let us now formulate our theory more precisely.1) Although the gauge particles are massless. These two problems can be solved simultaneously by going to lightcone coordinates: we write [8] p ±=_L(p3±p0)j ~ =(pl!p2). 1. Let us consider any large planar diagram (fig. Planar diagram theory 467 dual models. and the physical spectrum is governed by long range forces. (3. The propagators are then (2nfi (pz + 2p+p+ m2ie) (4.
6. divided into two regions a and b (see text).2) with respect to p ~.. The components p of the momenta of the propagators that cross the dotted line in fig. 1 0(xV)exp/— + (2TT)32P 2p+ (m2+p2) (4. Example of a planar diagram. 't Hooft. Planar diagram theory Fig. 5.3) tot Fig. (always directed anticlockwise). and to each window F?U) of the graph corresponds an integration over the momenta p.468 G. . 5. y> nate x. In terms of these variables the propagator is the Fourier transform of (4.^ and pp. + .
3). Here x + is the time difference between the two end points of the line. But it is easy to convince oneself that in those cases the diagram is zero as a consequence of the 8 functions. and (p.G. then we must expect lines going from (a) through (b) back to (a). >a. 7 will get several "sheets". <a. p j . For simplicity we shall assume that all external lines with positive p+ lie next to each other in the plane*.p+) are the difference of the momenta (p ( . If we divide the set of vertices into: (a) those w i t h * ? . ) circulating in the windows at both sides of the line. The parametrization problem can be solved by exploiting the famous 9 function in (4. 5 and have been numbered accordingly. and (b) those withxT . A new representation of the same diagram.. . Note that the propagator (4. The blocks here correspond to the propagators in fig.3) is Gaussian in the transverse momenta p. then all lines * going from (a) to (b) have t If this condition is not fulfilled the resulting plane of fig. etc. If we want to keep the diagram planar while dividing it into blobs (a) and (b). 7. 't Hooft. Planar diagram theory 469 10 1 1 1 2 14 15 13 16 1 8 17 19 20 24 25 26 27 31 32 33 23 28 2 1 22 29 30 tot Fig.
We see that the variables p + and x + are suitable coordinates. 7.5) 5. Now imagine a horizontal line with length p  o t a l and divide it into segments. and all amplitudes become real. (5. It is convenient at this point to perform a Wick rotation. T) is now a continuous variable on a similar rectangular surface [9]. Apt stands for the width of the block between / and/. together with the factors i at each vertex: (27r) 3 zXd. (4.1) and (5. and vertices are now horizontal lines.t + ^(27r) 3 Adr. loops in the original diagrams now correspond to vertical lines. Gaussian integrals (the 8 function in (4. (4. ix+ = T.3) now becomes 6 (rp+).1) we have a partition of the dual surface into meshes.3) now disappears. The integration in the transverse momenta (or coordinates) is Gaussian. in which the propagators correspond to blocks. The propagator is also Gaussian in terms of the x. and secondly in (5. The difference between (5. If we now vary the number a.2) where x (o.  * / ) • ( 5 J ) where C is independent of the transverse variables. 5. t Hooft. Planar diagram theory positive p . and with a length equal to the (positive) value of p+ in that propagator (fig.470 + G.4) The factor i in the exponent (4. or at each horizontal line in fig. then this line sweeps out a surface with constant width. we could study the diagrams in transverse coordinate space. each corresponding to a propagator going from (a) to (b).) ( * .1) one must . 7.r. Summing and integrating over all possible topologies in the p+x+ plane is equivalent to performing the remaining p+x+ integrations and the summations over the diagrams. and the summation is performed over all pairs of adjacent horizontal lines in fig. 6).2) is profound. See fig. defining the new regions of integrations). 5.Zy ij ^Ptj 2 (r. The integrand is (after the Wick rotation) C exp {. 7 in which we numbered the blocks corresponding to the propagators in fig. Comparison with the dual string Instead of considering the transverse momenta p. The first difference is that in eq. . This is to be compared with (the essential part of) the functional integrand for the quantized string: '«>/** [(gM) \2~ (5. Then we would have a transverse variable:? at each vertex of fig.
. Planar diagram theory All also integrate over all longitudinal variables and sum over all diagrams. . . If no vertex occurs betweenx\ and x£+l t n e n o n 'y ^o contributes to x k+l x k Taking v m . . 6: a number of "particles" is sitting on a line segment with length p^tal. with 4 + 1 . There exists.xt. x + . however. . . A Hamiltonian formalism Attempts to attain more understanding of the peculiarities of planar diagram field theory have failed until now. r = 0. we confine ourselves to the planar diagrams of 0 3 theory (again defined by means of a certain N > °° limit). This integration and summation together correspond to the summation over all partitions into meshes. .2) in any way.4 = e. . 't Hooft.1) can be approximated by (5. and from those it will probably depend whether (5. The x+ axis is divided into small segmentsxt.G. (6. a Hamiltonian for this system that might be useful. We put p total = 0. Expand e ietf = j _ ieH = x _ ie ^ + H ^ (6 3) where H0 will be taken to be diagonal in the abovedefined representation. 1. If so. so we can take x+ to be real. 6 actually correspond to loops in the original diagram).They have coordinates p *.—— 2(ptpt_) / . . r.A transverse loop integration momentum Pj is assigned to each particle (the particles in fig. 6. A representation of states i//> in a Hilbert space is defined as a set of structures like in fig. A Wick rotation is not necessary here. (6. so that p = 0 on the boundaries at the left and at the right. then the dual string will be an approximate solution of the dynamical equations of our gauge model.p . i = 1. Now we write the amplitude formally as A= + (6. . For simplicity. . The detailed structure of the meshes will depend on the initial Feynman rules. . We now construct the Hamiltonian H that will yield the sum of all planar diagrams. 2 + (p.1) (\jj\e'eH\\p) x < out I e'eH I \p > + x + <<// in> x + .2) xn n\ n~2 0 where summation and integration over the intermediate states is understood.4) H0 = IJ .
7 (7. (6. nt and X' as our elementary fermions we can again consider theN>°° limit. p* and pj are the coordinates of the closest neighbours at the right and at the left of the point p+. Here H^ is in action.3).1) Takingpt. guided by the topological structure of the dual theories. p+). then the dual coupling constant will be calculable and of order \. The X quark will then sit in the middle of a string with p and/or n quarks at its ends: we have a string with 2 or A baryons! Similarly protons. then the original particles will condensate into a string that keeps quarks together. with [at (p. thus giving rise to the required factor l/p + in the propagator (4. our particles are created or annihilated.5) creating respectively annihilating particles. Substituting this interaction Hamiltonian into (6.3) and (6. In that case we can raise or lower indices in the following way: \. It will be clear that in the case of baryons the 1 IN expansion is extremely delicate. If calculations will be possible at all in this theory. The HanNambu theory clearly suggests N = 3. with V (Pr ~Pi)(Pr ~P ) (P Pi) where X is the coupling constant. a (k. the situation is even more complicated. we are led to the planar diagram field theory. 7 always occurs twice. then its spectrum should come out to be discrete.X'1'. 7). But.472 G. we find exactly the Feynman rules for planar diagrams: the square root of the width of each block in fig. Conclusion We are still far away from a satisfactory theory for bound quarks. Let us define operators fft (p. 7. t Hooft. If our theory is to describe hadrons. despite the fact that the zeroth order Hamiltonian is continuous.3)). Planar diagram theory we get the correct exponential parts of the propagators (compare (4.2). k+)] =82(pk)8(p+k+). In our gauge theory model. We can then take //j = V+ F t . * X'>' = el'k \k = . neutrons and all other baryons can be constructed. p+) and a (p. As for baryons. p + ). . a similar Hamiltonian will describe one quark and one antiquark in interaction. At the vertices (horizontal lines in fig. in terms of which our problem can easily be formulated: if the eigenstates of a certain Hamiltonian crystallize into a discrete spectrum.
B. [9] J. Marseille Conf. Veltman. Nielsen and P. B64 (1973) 205. "DIAGRAMMAR". Phys. Gross. Gross and F. 1770 (1973). Rev.. [2] M. Nucl. Phys. 't Hooft and M. Phys. Virasoro. D. [6] K. [8] S. unpublished. Veltman. unpublished. Wilczek. Phys. Nambu. Mandelstam. 't Hooft. Phys. 139B (1965) 1006.D.J. Phys.Y.L. Lecture given in Orsay (August 1973).349 G. Rev. Nucl. 't Hooft and M. Gross and F. Letters 30 (1973) 1343. Phys. G.A. [7] H. B61 (1973) 455.J. B50 (1972) 318. Gervais and B. . Phys. Planar diagram theory References [1] G. Chang and S. Rev. Rev. Coleman and D.G. Rev. B62 (1973) 444. Ma.B.K. Sakita. Letters 32B (1970) 203. Letters 30 (1973) 716.J. Princeton preprint (1973). Rev. CERN preprint TH. NAL preprint. [4] G. Wilczek. 't Hooft. Letters 30 (1973) 1346.J. [5] G. Olesen. [3] D. Phys. Princeton preprint (1973). Han and Y. Testa. H. Nucl. 1923 June 1972. Letters 24 (1970) 1146. S. Wilson. Amati and M. Sakita and M. Phys. 180 (1969) 1506. Politzer. 't Hooft. S. D. CERN report 739 (1973).
It is clear that if two such mappings belong to different homotopy classes then the corresponding fields A^ andAJP cannot be continuously deformed one into another. Moscow. Our idea is to search for the absolute minimum of the given component of the phase space. A. From (1) it follows that \x) (5) 1 where K ^ a /sp(V 7 . The topological nature of the solutions is discussed. The space is euclidean and fourdimensional. Hence every field A^x) produce a certain mapping of the sphere S 3 onto the gauge group G. By "pseudoparticle" solutions we mean the long range fields A^ which minimize locally the YangMills actions S and for which S(A) < °°. All fields we are interested in satisfy the condition: F 3 A b A +[A . through the Euler angles. the phase space of the YangMills fields are divided into an infinite number of components. number 1 PHYSICS LETTERS 13 October 1975 PSEUDOPARTICLE SOLUTIONS O F T H E YANGMILLS EQUATIONS A. It is well known [2] that there exists an infinite number of different classes of mappings of S 3 * G if G is a nonabelian simple Lie group. since it is the invariant differential form of the appropriate dimension.A.S. Then the invariant measure will be: * Fonnulas like (3) are known in topology by the name of "Pontryagin class". TYUPKIN Landau Institute for Theoretical Physics. each of which is characterized by some value of q. (3) To prove this let us use the identity: tivky J. In the previous paper by one of the authors [1] tireimportance of the pseudoparticle solutions of the gauge field equations for the infrared problems was shown. USSR Received 19 August 1975 We find regular solutions of the four dimensional euclidean YangMills equations.AV through the field A^. d 4 x. In this case it is clear that: dn(g) = sP(g1dgXgldgxg1dg) (6) (7) is just the invariant measure on this group. Let ^(?i?2^3) b e s o m e parametrization of SU(2). say. POLYAKOV. The sphere itself is of course 3dimensional.S. In order to do this we need the formula expressing the integer q Now consider the case G = SU(2). Sp [F It is easy to check that F. Hence. (1) Consider a very large sphere S 3 in our 4dimensional space. 85 . Academy of Sciences. r nv Ky a a (4) ^spCVV^V^8irA From (4) follows: ' 0 . A. q=— e . SCHWARTZ and Yu.350 Volume 59B. The meaning of the notation in (7) is as follows. (2) A)i=g\x)dgldxti where g(x) are matrices of the gauge group. BELAVIN.4 g )d a a 3 9£W 6X. The solutions minimize locally the action integrals which is finite in this case. Let us start from the topological consideration which shows the existence of the desired solutions. In the present paper we shall find such a solution. where q is a certain integer.M.
K. One may consider of course several 86 .. =• ±.Akl). Comparing (8) with (5) we see that the integrand in (5) is precisely the Jacobian of the mapping of S^ on SU(2).r .v pv\y Fa& Xy' (13) (a are Pauli matrixes). PHYSICS LETTERS (8) 13 October 1975 3J. There exists an important inequality which will be extensively used below. For arbitrary group G one should consider its subgroup SU(2) for which A^ is given by (18) and all other matrix elements of ^4 let be zero.T 2 / 2 ) ( 5 + (f'lT+f2)(x + x„x 5 5 „ . From (9) and (3) it follows that: E>2nl\q\ where S(A)=E(A)/g2 andg 2 is a coupling constant. Let us search the solution of (13) which is invariant under simultaneous rotations of space and isotopic space. T)=—£ f(T) 2 r +X2 r (16) where@is an arbitrary scale. (15) . (3 p vaJ (9) (3 v pa FKy.df.V • :e 2 ±F' tiv\y \y are equivalent to the following one: e a(Sy& FyS = e p. two equations: F' p.5 „5 ) x 5 _x x 5 . In the case of the arbitrary group G one should consider the mapping of 5 3 on its SU(2) subgroup and repeat the above.x 5 ).e 2 a(3y6 F yd F = 3 A. We do not know whether any solutions of (13) exist with q > 1... contains the arbitrary scale X. bRA + [A A J (11) Again it is sufficient to consider the case G . Hence these fields are long range and are essential in the infrared problems. which replace the usual YangMills one: F a(J It is evident that the first tensor structure (15) satisfies the equation (13) and the second does not. In this case it is convenient though not necessary to extend this group up to SU(2) X SU(2) « 0(4). number 1 dfi = Sp [g 9$. as is evident from the scale invariance.. It is just the definition of the mapping degree. Comparison of (17) and (10) shows that we find absolute minimum for q = 1. We shall show now that for q = 1 this bound can be saturated.+ixa)(xUx2)ll2 T2=X2+X2 ±A' =UA°'±U. In other words one can search the solution of the equation.351 Volume 59B. Hence we are to choose: (10) flr+r=0. The formula (10) gives the lower bound for the energy of the quasiparticles in each homotopy class. (12) Now. The SU(2) gauge field are connected with/l a ' 3 by the formulas: •\x) dg(x) dx (18) (x. d$.x. Our solution. The quasienergy E is given by ^ i S p / ^ d 4 * (17) 2 4 2 = ^ S p / ( ^ ) d x = 2. Consider the following relation: Sp f(F F ) d x>0 2 4 (14) It is easy to calculate F: /^(3 = ( 2 / .SU(2). . Hence q is the number of'times the SU(2) is covered under this mapping. The only possibility is: Aa& =/(T)(X 5 „x„5 ). The gauge fields for 0 ( 4 ) are AaV where A are antisymmetric on a3. Another representation for the solution (14) is given by the formulas: A = " r 2 + X2' g(x) = ^£=1.
Novikov for explanation of some topological ideas. References [1J A.P. 59B (1975) 82. we do not know whether they are attracted to each other and form the pscudoparticle with q > 1 or whether there exists repulsion and no stable pseudoparticle.P. number 1 PHYSICS LETTERS 13 October 1975 pieudoparticles with q = 1.352 Volume 59B. Lett. However. Phys. Polyakov. One of us (A.M.) is indebted to S.M. 87 .
excluding the time components A°&. Having in mind a YangMills theory. Specifically in a theory of fermions coupled to YangMills fields. Massachusetts (Received 1 June 1976) We propose a description of the vacuum in YangMills theory and arrive at a physical interpretation of the pseudoparticle solution and the attendant violation of symmetries. Jackiw and C. and Tyupkin. NUMBER 3 PHYSICAL REVIEW LETTERS 19 JULY 1976 COMMENTS Vacuum Periodicity in a YangMills Quantum Theory* R. symmetrynonconserving effects a r e found through the pressence of the axialvectorcurrent anomaly. 3 Thus he provides a possible resolution of the longstanding U(l) problem 4 and an intriguing suggestion for the origin of CP nonconservation. But it_is important to realize that there are values of A that can be obtained from each other by . where g is the gauge coupling constant. i. The phenomena are 0(exp(. (1) where g is the unitary matrix of a gauge transformation that can be joined to the identity through a oneparameter continuous family of transformations g(x. one does not integrate over potentials of the form l&=g1(K)Wg(x).e. (2) The potentials^of Eq. Schwartz. parametrized by a CPnonconserving angle. The fact that the classical field configuration which is responsible for the new results is in Euclidean fourdimensional space. Without repeating details of the wellkown gaugefixing procedure.l)=g&. with chiral U(l) and CP symmetries. which we hope. 1 with the suggestion that it be used to dominate the functional integral which describes a quantum field theory continued to Euclidean space. In particular. a): gfr.8tr2/g2)). 't Hooft2 has shown that these nontrivial minima of the action give nonvanishing contributions to amplitudes which would be zero in the ordinary sector. will clarify the physical interpretation of the pseudoparticle solution and will supplement 't Hooft's more formal 172 computations. Rebbi Laboratory for Nuclear Science and Physics Department. Here we wish to present a further explanation of this point. (1) are of course gauge equivalent to A = 0. In defining scalar products and matrix elements of observables one must avoid infinities associated with the volume of the gauge group..0)=I. A classical pseudoparticle solution to the SU(2) YangMills theory in Euclidean fourdimensional space has been given by Belavin. g(x. The existence of topologically inequivalent classical gauge fields gives rise to a family of quantum mechanical vacua. let us only recall that it removes from the functional integral over A configurations of the fields which can be joined by a continuous gauge transformation to configurations already counted. they are nonperturbative. we have taken the potentials A(x) (antiHermitian matrices in the space of the infinitesimal group generators) as argument of the functional. Polyakov. Our considerations lead to a description of the quantum mechanical vacuum state of a YangMills theory which is unexpectedly rich. Cambridge. a state of the system can be represented by a wave functional *[A] of the field configuration. In the quantum field theory. 5 Also the exponentially small magnitude is indicative of tunneling. The requirement of vacuum stability against gauge transformations renders the vacua chirally noninvariant. because they are dependent variables. imaginary time.VOLUME 37. leads one to suspect that the pseudoparticle is associated with quantummechanical tunneling by which field configurations in the ordinary threedimensional space a r e joined in the course of the (realtime) evolution through the penetration of an energy barrier. Massachusetts Institute of Technology.
To exemplify this. Now the field strength is nonvanishing. with all the directions at <*> identified. Such an evolution should be a r °ntinuous gauge transformation. should not be r e moved from the integrations over the field configurations by the gauge fixing procedure. In the quantum theory. The energy. Into the group space.x2) + 2x(a • x) + 2Ax xa] •uid of course to vanishing field strengths F{j. __ _ But the pseudoparticle solution connects An with A n+1 . so thatg' defines a mapping S3TS3. but not continuously ( . We can make contact now with the pseudoparticle solution. Neglecting tunneling effects we might expect the vacuum to be of the form ^„[A]=<P[AA„]. Before proceeding. From the topological point of view. (10) To determine the coefficients c n in this equation let us observe that the finite gauge transformation gl changes ipn into j/* n + 1 . we require i'(x) 2t'X (4) M. the Euclidean space E3 with points at « Identified is equivalent (homeomorphic) to a threedimensional sphere S 3 . The physical implication of the pseudoparticle solution is that the quantal description of the vacuum state cannot be limited to fluctuations around any definite classical configuration of zero energyLet us now describe in greater detail the nature of the vacuum wave functional. 5 The pseudoparticle solution 1 serves precisely this purpose: It carries zero energy (the Euclidean s t r e s s tensor vanishes). i. Values of the potentials like_ those of Eq. Consider any of the field configurations A„(x)=g„1(x)Vg„(x) (8) It is known that these mappings fall into homotopy classes (mappings belonging to different classes cannot be continuously distorted into each other) classified by an integer n. (4) by \ . NUMBER 3 PHYSICAL REVIEW LETTERS 19 JULY 1976 transformations which cannot be continuously joined with the identity transformation. let us mul'i. All paths joining the " v ° field configurations in real time must go over •tn °nergy b a r r i e r . although gauge equivalent to A = 0. It is well known that a semiclassical description of tunneling can be given by solving the classical equations of motion with imaginary time. (4). .a 2 ./. potentials.'y the potentials of Eq. The true quantal vacuum state will therefore be a superposition of the form *[A]=£„c n ! UA] + 0(exp(8. (6) a adiabatically from . but this is impossible because gt and the identity belong to dif'°rent homotopy classes. thus achieving an evolution which would be classically forbidden for real time. and indeed we shall argue that physical effects are a s sociated with them. it can be arranged to connect g =gl at x 4 = .a and increase where the wave functional <p is peaked about zero and has a spread due to quantum fluctuations and any A„ can be chosen as representative of the classical vacuum.. but proportional to ..354 VO IUME 37. (9) C O ' iri given in Eq.T 2 /ir 2 )). and that 'here is no energyconserving evolution of the system which adiabatically connects that configuration with A = 0.\ to i .\ to +j. (3) being a representative of •he nth class.iugeequivalent. (5) Thus g defines a mapping of the threedimensional space. but the manifold of SU(2) is also homeomorphic to S 3 . when we consider a gauge transformation £ .it=*> with g=I at %4=°°. tunneling will occur across this b a r r i e r . Requiring the vacuum state to be stable against gauge transforma173 . the classical zeroenergy configuration.? ) 2 and exhibits a barrier shape as a varies from . . 1 Observe that the field configura'•on of Eq. becomes proportional to ( a 2 .e. For instance. let us characterize the classes of gaugeequivalent. A'n(x) = lir1(x)]" v th with vanishing F*'.j \d3x TrFuFu»0. We study effects which are local in space and therefore. we may consider 2 A2 i'^FTx2" 2tAcrx x 2 + A2 (3) which gives origin to A(x) =g{ '(xJVg^x) = ( j p + x 2 ) 2 [a(X2 . (4) has zero potential energy./)„. giving origin to tunneling between the different .
(11).) The fact that Q5 commutes with the Hamiltonian. R. Ann. Lett. D. Belavin. 6 A nonzero value of 0 could describe CP nonconservation. (g belongs to the nth homotopy class when it is continuously deformable tog„. M. and D. Phys. N. We also thank S. and D.x) 174 (16) . 82 (1975). Mass. We a r e happy to acknowledge our indebtedness to G. Chem.2 e M U « S may be given. 343 (1976).) 13. Coleman and L. D 13. NUMBER 3 PHYSICAL REVIEW LETTERS 19 JULY ly. Treiman. whereas in the massless case conservation of Q5 renders all vacua degenerate. G J. tions determines the coefficients to be c n = e<"6. P r e s s . K. Added note. J. we expect that if the fermions are massive. Math.) 68. Swieca.. Dash en. *A. Phys. Grisaru. but gaugevariant. where under the gauge transformation Si *e[A]rieie*e[A]. 4 H. Energy Research and Development Administration under Contract No. parametrized by an angle 0. Susskind. a pseudoparticle solution exists in Euclidean twodimension space—it is just the NielsenOlesen string To exhibit the gauge dependence of Q 5 . 59B. 56. in Lectures on Elementary Particles and Quantum Field Theory. current is given by J5M = J 5 M _ 4 7 r .Y. and Y. Coleman. (12) The occurrence of multiple vacua is intriguing and is reminiscent of the situation encountered in the Schwinger model. 1099 (1972). we received a paper by C. A. Gross (to be published) who arrive at conclusions similar to ours. Schwinger. Polyakov. Equation (15) also demonstrates the possibility of symmetry breaking without Goldstone bosons: This may provide a solution to the U(l) problem. Jackiw. Gross (Princeton Univ. 1972). and H. Phys. Schwartz.[A] (15) which in turn shows that all the vacua a r e degenerate in energy and define the same theory. Phys.355 VOLUME 37. (N. A. However. by adding to the model charged spinless fields.. Lowenstein and A.. But Q5 is conserved.3 0 6 9 . Rev. Callan. 2 . 2 G. Jackiw. Princeton. Freed. S. 2 but in the theory a s developed thus far there is no indication how to compute 0.—After completion of this manuscript. 2425 (1962). 't Hooft. we perform a finite gauge transformation g. 5 Use of imaginary time in discussions of tunneling is well known. 3 A conserved. R. 1970). One may then introduce the U(l) axialvector current which however is not conserved because of the anomaly. Pagels. so that Q5 is no longer conserved. Phys. 6 The significance of the phase 9 in Eqs. J. and lectures at various universities. R. 692 (1972). see Eq. Susskind for discussions. Tyupkin. (13b) *Thls work is supported in part through funds provided by the U. implies that e x p ( . R. Rev.l ) . Phys. ^28. Cambridge. d3xTreijk(giaig)(gidjg)(gldkg) (14) where df±(g) i s the invariant measure of the group and n is the integer which characterizes the homotopy class of g.2 . (N. Finally. edited by S. E ( l l . with a Higgs potential and minimal electromagnetic coupling.. (11) Thus we find a family of vacua. see. and therefore AQ5 = . 172 (1971). 59B. together with Eq. Ann. Adler. (14).g. xTr(AT8aAB + iAvAaAB) (13a) and the conserved axial charge is Qs = Jd*xJs°. 't Hooft. 8 (1976). and find ^Q5=Y2^j = j2^jdlx(g)=2n.) 93. Phys. J .) There is no pseudoparticle solution for spinor electrodynamics in two spacetime dimensions. A. and L. edited by S. 85 (1975). Lett. Jackiw. different values of 8 define nonequivalent theories as in the Schwinger model. we remark that. (This analogy was developed in conversations with Coleman and Susskind.Y. in Lectures on Current Algebra and Its Applications.. L. S. Deser. (10) and (11) becomes apparent when massless fermions are coupled to the YangMills fields.Y. whose calculations made our observations possible. Lett. 267 (1975). S.i t e ' Q s ) # e [ A ] = * e + e. McLaughlin.4 An explanation of the nonconservation of the gaugeinvariant fermionic axial charge Q5(t) = fd'xJ5°(t. with g(x)^~I. (N. J. Phys. Pendleton (The MIT P r e s s . e. and that it changes by two units under the gauge transformation#!. Phys. Polyakov. 37. Rev. The tunneling process between adjacent vacuum components ^„[A] and $„+1[A] is equivalent to a gauge transformation g1 which changes Q5 Q5 by two units.
1975). and e ^ F " " is also p r o " tional to the anomalous divergence of the axial vec tor current. Zichichi (Academic. New York. K. R. 45(1973)]. . in Laws of Hadronic Matter. NUMBER 3 PHYSICAL REVIEW LETTERS 19 Ju j phys. Johnson.356 E 37. edited by A. Phys. Lett. 5. 253 (1963). The topologically conlervc'd object is Jd2xellvF>lv. B61. Jackiw.
J2 is the Langrange density of the theory. When massless fermions are present.)+ £g{]} (1) * Research supported by the National Science Foundation under Grant Number MPS 7522514. Princeton. In nonabelian theories of the strong interactions one finds spontaneous symmetry breaking of axial baryon number without the generation of a Goldstone boson. DASHEN* Institute for Advanced Study. USA R. a mechanism for chiral SU(A0 symmetry breaking and a possible source of T violation. in effect. Princeton University. E(lll)2220. To explore the structure of the vacuum we study the Euclidean functional integral <0exp(#OIO> . each one of which is a superposition of the vacua with difinite topology and stable under the tunnelling process. USA and D. The construction and properties of these vacua are analyzed. P and T are spontaneously violated in some of them!). shown that the vacuum "seizes" as suggested by Kogut and Susskind [2]. In the spirit of displaying qualitative consequence of the new vacuum structure we shall also briefly summarize results obtains! from rather crude approximations to the functional integral. As pointed out by 334 .J. inequivalent worlds (most striking.. however. we find a continuum of vacua. been obscure since they are localized in time as well as space. G. New Jersey 08540. The physical interpretation of these solutions has. Now since F^ = 0 implies A^ = g 1 (. In fact. CALLAN. number 3 PHYSICS LETTERS 2 August 1976 THE STRUCTURE OF THE GAUGE THEORY VACUUM* C. The new vacua are the ground states of independent. Princeton. Research sponsored in part by the ERDA under Grant No. New Jersey 08540. any gauge field included in the functional integration defines a map of the sphere at Euclidean infinity into G. Diagonalization of the Hamiltonian then leads to a continuum of vacua. and identified the mechanism by which it does so. Joseph Henry Laboratories.. New Jersey 08540. Our primary aim in this letter will be to give arguments for the existence of the new vacuum structure and to present the correct form of the functional integral appropriate to studying the properties of a particular vacuum. In this letter we shall show that Euclidean gauge solitons describe events in which topologically distinct realizations of the gauge vacuum tunnel into one another and that this process radically changes the nature of the vacuum state. Prinecton. USA Received 20 May 1976 The finite action Euclidean solutions of gauge theories are shown to indicate the existence of tunneling between topologically distinct vacuum configurations. and in general. Jr.G.357 Volume 63B.^ JlDA^DiP] Xexp{fddx[£(A^<Pi. Polyakov [1] has recently pointed out that the Euclidean classical equations of motion of gauge theories have solitonlike solutions and has suggested that when properly included in the Euclidean functional integral they may have a bearing on the dynamics of confinement. GROSS Joseph Henry Laboratories. i? f is a gaugefixing term and the integration is to be done over all fields that approach vacuum values (F = 0) at infinity.t)9 M g(x) takes on values in the gauge group. Princeton University.F. We have. the vacuum tunnelling process forces a redefinition of the fermion vacuum as well and leads directly to spontaneous breakdown of chiral invariance without generating a "ninth" Goldstone boson. where d is the dimension of space time.
Then since v is integral the gauge vacuum configuration at t = — °° must also have integral winding number «(+ °°) = n (—°°) + vTherefore we must admit the existence of a discrete infinity of vacuum states. (1) fall into discrete classes indexed by an integer v running from — °° to + °° (we shall use the notation IA4„] v to denote functional integration over the fth class). is the "right" one. The dynamical variables are now just the space components. is then straightforward: ^MHt)\m) ^ J{DA^. In these theories (the only ones we consider) the gauge fields integrated over in eq. For most nonAbelian groups in four dimensional space time and U(l) in two dimensional spacetime. rij_j(G). of course. all our vacuum states \n) are annihilated by such local gauge transformations.358 Volume 63B. Since Tis a gauge transformation.=/dd1x[F0Wa+*W where Dt is the covariant derivative and /Q is the gauge source of fields other than the gauge field itself. the hamiltonian 335 T^ e Hki^ x tr <Ai Aj Ak). gauge invariant). In the v ¥= 0 sectors the minimum action is in general nonzero — for v = 1 in four dimensions it corresponds to the Belavin et al. associated with any particular Euclidean gauge field time history may be written as a gauge invariant volume integral 1 functional integrals corresponding to different yclasses. In order to satisfy Gauss' law. perturbation treatments of gauge theories expand about A = 0 and pretend that the vacuum \n = 0> is true vacuum.. The interpretation of the multiplicity of Euclidean . v = n(t = + °°) — n[t = — °°). and T satisfies T\n) = \n + 1>. instanton [3]. labelled by a winding number taking on integral values from — °o to + °°. this is completely wrong and causes perturbation theory to miss qualitatively significant effects. The topological quantum number. it is sufficient to restrict the state space by Q^ty) = 0 for all gauge functions X which vanish at infinity. where n= GJ. Thus in the same WKB sense the \n) *• \n + 1 amplitude is > 0(exp(87r2/g. Thus there is actually a discrete infinity of funtional integrals and one must ask which. [3] these maps fall into homotopy classes corresponding to elements of the homotopy group. This is a typical "tunnelling" amplitude.](n_m) (4) Xexp{/ddx[i?(^)+ £gf]}. V (this boundary condition is. if any. whose action is 87: /g . which change the vacuum topology. At. number 3 PHYSICS LETTERS 2 August 1976 Belavin et al. with Gx (2Tr/g)[E(<x>) + E(. for the vector potential and at large negative and positive times they must take on time independent vacuum values. but finite. In particular. d= 4 (2) / * • _2 4n •*vf""' d = 2.= gJ$. The functional integral over homotopy class v describes a vacuumtovacuum transition in which the vacuum winding number changes by v\ Now the minimum action for v = 0 is zero (corresponding to A = 0) so that in the WKB sense the \n) »• \n) amplitude is 0(1).. the integrand is a total divergence and ^40 = 0 gauge v may be rewritten as a surface integral. Indeed. In both cases. vanishing exponentially for small coupling and unseen by standard perturbation theory. T.). Aj(x) = g~i(x)dig(x).»)] for the twodimensional abelian theory or £ „ = (2n/g) / d2StEfx a for the fourdimensional nonAbelian theory. There also exist gauge functions which do not vanish at infinity and generate gauge transformations. What then is the true vacuum? A convenient way of constructing it is to consider the generators of time independent gauge transformations characterized by a gauge function Aa(x): jY*tr(F^). \n). this homopoty group is Z. spacetime volume. lJ^ "£/* With no loss of generality (we have the freedom of making time independent gauge transformations) we may choose n(t = — °°) to be an integer. JD/FQ. One can clearly see what is going on by working in the gauge A0 = 0 and requiring F to vanish outside a large. Because of vacuum tunnelling. v. d = 4 (3) d = 2. One can easily construct a unitary T effecting such a nonlocal gauge transformation: 7"= exp(iGx).
0 } (V n+ln_\ V Z) n+. In the path integral this corresponds to performing ordinary perturbation theory about the appropriate classical solution for each topologically distinct sector and summing.itxp{fddx[£g{+ £gi] (5) where £e = (i0/87r2) tr (F^ FM") for d = 4. Having constructed a vacuum one can then calculate Green' functions of gauge invariant operators perturbatively. If we try the above sort of approximation on the nonabelian theory in four dimensions. £e = (Wj4ir) e F for d = 2 and in the second expression for 1(6) all gauge field topologies are summed over. e x p { i f l ( » + . its eigenvalues are e1 . and the eigenstates are 10> = 2 e ' " e ri). The second form for 1(6) makes manifest one of the peculiar ways in which the 6 worlds differ from one another.. By contrast. the 0 vacua have an energy per unit volume equal to —2FQ~ COS 6 e °. neglecting any interactions between vortices (since fields decrease exponentially this is not too bad for low vortex density. Although the 0 = 0 vacuum has lowest energy (and no parity violation) we can't conclude that it is the vacuum since the other 0vacua.. as well as an arbitrary position.. 1(6) contains all possible information about physics in the 0world and requires no further modification.( " + + O S .. this energy difference is a nonperturbative effect (a tunnelling effect) but potentially important nonetheless. In this "dilute gas" approximation. This clearly breaks P and T invariance (except for 0 = 0) and we must in general expect spontaneous breaking of spacetime symmetries in all but a few special 9worlds! As a concrete illustration of these general remarks we should like to present an approximate evaluation of 1(6) in twodimensional charged scalar electrodynamics. SQ. The integration over X need not diverge since scale invariance is broken by quantum corrections. It is gratifying that this multiplicity of worlds is known to exist in the Schwinger model. } (6) <e'l exp(#oiei> —>• 6(o . Because SQ<* \/g . each 0> vacuum is the ground state of an independent and in general physically inequivalent sector within which we may study the propagation of gauge invariant disturbances. The sum is trivial and yields exp [2(V/V0)e ° cos 6). In the sector with v = ± 1 the field configuration with minimum action is just the NielsonOlesen vortex [5 ] in which there is a localized region of non where the factors of V come from integrating over vortex locations and VQ is a normalization factor which can be calculated from the quantum corrections to this basically semiclassical approximation. proportional to \r \g .. F H . the effective Lagrangian is \x[FyilF. the functional integral is <0'exp(#f)0>~6(0'0) .=0 exp { . In the dilute gas approximation one finds for the vacuum energy density 336 . The classical theory is scale invariant and the basic v = 1 solution (instanton) has an arbitrary scale parameter. In four dimensional pure YangMills theory. reexpressed in Minkowski coordinates. Thus.J V * [ i ? 0 4 M .j„ e x p { ..w + ( 0 / 8 T T V M ! . Since the different 0worlds do not communicate with each other. the renormalization group should tell us whether the integral converges at the short distance end. there is a problem. We have normalized the energy so that the naive perturbation theory vacuum energy is zero.) + i? gf ] } = f[DAli. Finally. (4). This diagonalization of H is obviously unaffected by including in 2 sources coupled to gauge invariant densities. there is no apriori way of deciding which world is the right one. which turns out to mean small g).e'y(e) 1(6) = Z/exp(u>0) V x / [ D ^ M . Indeed. number 3 PHYSICS LETTERS 2 August 1976 commutes with it and energy eigenstates must be T eigenstates. eq. corresponding there to different values of background electric field [4].359 Volume 63B. Since 7" is unitary. We shall construct the sectors with topological quantum number v by superposing n+ v = + 1 vortices and n_ v = — 1 vortices with n+ — n_ = v. 0 < 0 < 277. though higher in energy are stable to gauge invariant perturbations. arbitrary location and total action. in 0 basis: zero field of flux ±27r/g. X. radius n (pi is the heavy photon mass).n. we must express the functional integral.
Therefore. we may replace eq. There will. is not invariant under global gauge transformations.. color trf^e^^O^+l^. of vacuum states characterized by a vacuum gauge field with winding number n and a standard fermi vacuum with all negative energy states filled. Thus the attempt to construct the vacuum may run into an essential strong coupling problem because the quantum corrections to vacuum tunnelling will be large for large instanton size.g. 't Hooft [6] has noted that if A is taken equal to the v = +1 or v = — 1. If the vacuum V states of different topology are defined by \n) . In fact.Sn <. flavor. (4): Doing the fermion integrations for fixed A yields [det(jS .2 cos 6 J 0 X3 d X exp [". for massless quarks. This phenomenon is typical of a theory with no inherent mass scale which produces masses dynamically. affect the standard applications of asymptotic freedom which rely on one's ability to compute operator product expansion coefficients at short distances. On the other hand. (4) by (n\e~Ht\m) •+8„mf[DAll. can be computed perturbatively. however. (n e '\m) < bnm ! * The reason for this is that. We again confine nonzero F to a large but finite spacetime volume. In principle we must allow for transitions between vacuum. we must find (n\t~HtDv\m) oc 5 n _ m v where Dv is any operator of chirality 2Nv {Dv may stand for multiple insertions of local operators at different times — all that matters is net chirality)... 4 y/Cg). the 337 . Precisely because of asymptotic freedom. number 3 Cg = . If the theory is asymptotically free and there are not too many quark multiplets. the conserved axial charge where g(\/n) is the usual effective coupling. so that it must be true that (n\e~Ht\m) °= sn m. with Q510> = 0 one finds that Q5\n) = 2Nn\ri>. ji is arbitrary.. and evaluate <«e f\m) for general n and m. normalized so that g (X = n) = g. If one sets the renormalization mass scale. theng is determined (dimensional transmutation) and typically of order 1.7"" 0>. no matter how small one makes g. and again encounter a discrete infinity. there may not be a sensible way of perturbatively calculating even Green's functions of gauge invariant operators.)} U27T 2 J (8) while invariant under local gauge transformations. . and our argument is just telling us that whenever A^ belongs to a v =£ 0 class.360 Volume 63B. vacuum tunnelling is suppressed at arbitrarily small scales and leading short distance behavior will agree with conventional calculations. These problems should not.4)]+1 • This determinant vanishes whenever (i — 4) has a zero eigenvalue. in the limit of large instanton size. which reflect the mass scale set nonperturbatively by the tunnelling phenomenon. one is driven to large coupling (unless (3 has a small infrared fixed point) and the dilute gas approximation breaks down (instanton overlap and have long range interactions). V. The arguments presented above require some modification when massless fermions are present. The restriction on the topology of the gauge field histories would actually have emerged directly from a mindless application of eq. and 7summarizing the effect of loop corrections. Q5 is conserved.. however. /J. which changes gauge field winding number by one unit and T is the number of flavors. {!«>}. equal to some physical mass (e. because of the anomaly. However. ) + i? g f ]} (10) with the same meaning still attached to Dv. eliminating the v ¥= 0 sectors from the integration.)+£tf]} and (n\emDv\m)^8n+vm {[DA^. In particular.. In general. one has TQ$ T~l = Q5 — 2N where T is the global gauge transformation. 0 — 4) has a zero eigenvalue. be calculated nonleading terms suppressed by powers of momentum. This condition is met for any pure SU(7V) gauge theory and for SU(3) with no more than ten flavors of quark. . Though vacuum tunnelling is now suppressed.gWv) PHYSICS LETTERS +9(^(Xlu)) yj 2 August 1976 In fact.]0exp{fddx[Je<All. introduced earlier./ d ^ [ i ? U M .]^ X e x p { . instanton there is a zero eigenvalue.g can vanish rapidly enough for the integral to converge in the limit of large X (small instanton size).
The variation of the vacuum energy with respect to 6 is just <tr F F'"').fddx[£(A^. sum over n+ and integrate over instanton locations. tunnelling does have a profound effect on the vacuum energy and other physically relevant quantities. but gaugevariant. U(l) charge takes one out of a given 0> sector (eiaQs 0>= 0 + 2Na>) while tr(FF). we shall compute the determinant of 0 — jf) in the subspace spanned by the 2« + functions i//0 (X.. the integration over fermi fields yields det($ —4). Finally.361 Volume 63B. P and T appear not to be spontaneously violated and.xi ). fermion pairs are produced.. Now.. these functions are orthonormal and one has to compute the determinant of the matrix In this approximation A differs from a gauge transformation only in the neighborhood of each instanton and My can be approximated by where 0O = fa0. The solution to this problem is obvious (it was solved in the Schwinger model years ago!): The proper vacuum states are the \9) vacua. consisting of those elements which rotate 9 by a multiple of 27T. the divergence of the gauge invariant axial current. Since the interesting physical effects arise precisely from these zero energy solutions. Although the presence of zero mass fermions suppresses vacuum tunnelling in the strict asymptotic sense. That Q5 causes transitions between different vacua is characteristic of the "vacuum seizing" mechanism postulated by Kogut and Susskind [2] while the nonvanishing of FF in instanton solutions as a possible escape from the U(l) problem was noted by G. There is no associated Goldstone boson because the conserved. in which basis the functional integrals have the form <0'e"f0> W6)f[DAll. Although the pair must ultimately be absorbed by an antitunnelling. the axial baryon number invariance of the original Lagrangian is violated by a vacuum expectation value of operators with nonzero chirality. To get some notion of what goes on it is instructive to attempt a crude calculation of the basic functional integrals of eq. We shall assume that the integral over A is dominated by configurations of widely separated instantons (n+ in number) and antiinstantons (n_ in number).]{0) £#]} Xexp{fddx[j3(Ati. individual instantons have a zeroenergy eigenfunction \pQ±(x. To compute the vacuum energy we must set n+ = n_ (configurations with v = 0).y  as required by cluster decomposition and the vanishing of the "vacuum" expectation (n \D\n): it obviously approached (n\D+\n + l)(n + \\D\ri). The I//Q fall off at large distances. indeed. (11) in the case of a single flavor.) The cluster problem is resolved by the nonvanishing vacuum expectation value of D in the true vacuum state. For a given gauge field configuration. The 7 5 structure of the \p0 forbids .. The arguments of the preceeding paragraph show that (n\D\n) = 0. only broken axial U(l) and not axial SU(/V).' and / to be both instantons or both antiinstantons... the quantity whose nonzero value is the signal for P and T violation. Consider an operator D of chirality 27V. When the vacuum tunnels.. In the widely separated instanton approximation. has nonvanishing matrix elements. is independent of d. We will ignore the integration over instanton sizes. It should be said that at this stage we have in the /Vflavor case.. (11) X exp{.x±) (x± is the instanton location and the ± label distinguishes instanton from antiinstanton). all the \6) vacua are physically equivalent. while nonzero. a discrete subgroup of order IN of U(l) is left unbroken. 't Hooft [7]. In the massless fermion case.This determinant must also be approximated. which is why we introduce the 0O 's which are localized as well as the instanton itself. number 3 PHYSICS TTERS 2 August 1976 \n) vacua are not acceptable because they violate cluster decomposition for operators of nonzero chirality .exactly like the free fermi propagator. The 338 . Actually.x~[).)+ {d'\tHtDv\B)  b{d'6)J[DA)1. Then (n\D+(x)D(y)\n) will not vanish for large \x . since the fermions are massless the pair may live for a long time and tunnelling occurs freely in intermediate states.](v) /£gf] }DV . as 't Hooft has pointed out [7]. since the axial charge rotates 8.(n + \\D\n) =£0. The fact that only one topological class of gauge field history contributes to each functional integral makes physical quantities have a trivial dependence on 6: The vacuum energy density. i//^(x.
quantum chromodynamics. As a result. Integrating over instanton locations and then over the scale size X . Gross) would like to acknowledge V. if not impossible. Summing over numbers and locations of instantons simply completes the vacuum fermion loop analogy by providing all possible insertions of the pseudo mass terms. In the dilute gas approximation the energy difference is proportional to the integral over all instantons which overlap the quark. or momentumdependent mass term. Then calculations of physical quantities proceed in a perfectly conventional way so long as we remember to add the mass term ~ e~Sci(V+ + V_) to the massless quark propagator. and one is again confronted with a strong coupling problem. much remains to be done to fully exploit the phenomena we have found. We have constructed a simple HartreeFock type argument for N= 2 which has a chance of being correct in a weak coupling theory and which seems. but does something more complicated. In terms of the picture presented here Polyakov's ideas about confinement appear as follows. where Sc\ is the instanton classical action. of a vacuum that breaks chiral symmetry and sets a dynamical mass scale. Perhaps superunified theories will shed some light on these questions. The major difficulty. In the "scale invariant" four dimensional theory (tpip) ¥= 0 implies spontaneous generation of mass and dimensional transmutation as before. For the large instantons (small X). The inevitable Goldstone bosons arise in this case from iterated bubble graphs generated by the fourfermion interactions. The 7 5 structure of 0Q is such that V~ is like i//(l ± 75)1// in its Dirac matrix structure. For an isolated quark located at x the tunnelling amplitude \n. We are also intrigued by the natural appearance of spontaneous violation of P and T invariance but have so far not seen how to understand why these effects are small in the real world or how to exploit them to explain observed violations of these symmetries. V±.x)~* \n + 1. on superificial examination. The limit of zero mass is smooth. will be proportional to e cl and will have only the by now familiar dependence on g characteristic of a tunnelling process. We are especially encouraged by the appearance. it acts as a small perturbation on the /«Q = 0 theory. the dilute gas approximation is not valid. On the other hand. Anything which directly depends on the mass term. V* = 0o±(c)0n+(•*')' associated with the instantions.1 leads to an integral which tends to diverge at small X. Obviously.75) \pj). as it should. is basically a strong coupling theory and reliable calculations are difficult.J. Indeed. such as the vacuum expectation of t//i//. A quark state will then have more energy than the vacuum. However. which has the obvious interpretation of massless quark propagators connecting nonlocal vertices. the above discussion is modified in an important way: the effective instantonquark interaction is no longer billinear. One of us (D. We do not wish to make too much of these crude arguments other than to suggest that the new interactions generated by vacuum tunnelling are likely to play a key role in generation of quark masses and Goldstone bosons. the instanton (antiinstanton) vertex has the structure V+ = I l & f y l + 7 5 ) ^ (V_ = I l & f y O . one must break the global chiral SU(7V). on massless quark loops. one may hope that a new understanding of the qualitative physics will suggest new methods of calculation. Gribov for stimulating conversations and in par339 . number 3 PHYSICS LETTERS 2 August 1976 cycle expansion of the determinant then gives a sum of terms with a graphical interpretation in terms of closed fermion loops. For instance. summing over instantons in the dilute gas approximation will not just produce a mass term in the quark propagator. the effective interactions between quarks of different helicity generated by the instanton provide new ways of identifying sums of graphs which can lead to the desired symmetry breakdown. If the bare mass is small compared to the spontaneously generated mass. The functional integral weights each vertex with a factor proportional to e~ Scl . while we have argued that the vacuum tunnelling phenomenon is only guaranteed to violate the chiral U(l) symmetry. To produce a quark mass term. already in semiclassical approximations. however. but 2iVlinear. which is to say that V± looks like a nonlocal.x) will be reduced relative to the vacuum to vacuum amplitude. If there are N flavors. Vt. to generate quark masses. If the fermion is given a bare mass tunnelling is allowed and one is driven directly to a 9 vacuum (whose energy depends on 6).N.362 Volume 63B. of course. if n+ = n_ = 1 we get Det' <(<p+0\(trl\<t>o)(<Po\(t>r1\<Po). is that the theory we are really interested in.
Harvard preprint (1975). S.363 Volume 63B. 't Hooft. Lowenstein and A. Coleman. References [1] A. Rev. A. S. Rev. Polyakov. Phys. [3] A. S9B (1975) 82. [2] J. A. Belavin.M. [4] J. Phys. Dll (1975) 3594. Adler.S. B61 (1973) 45. Polyakov. Ann. Lett. Phys. Harvard preprint (1976). Lett. 59B (1975) 85. Phys.S. Nucl. Schwartz and Yu. 177 (1969) 2426. (5] H. Susskind. Jackiw. Phys. Swieca. number 3 PHYSICS LETTERS 2 August 19 ticular for the suggestion that the Euclidean solitons might be relevant to the structure of the vacuum. 340 . Kogut and L. Phys. [6] T. Nielsen and P. S. [7] G. Olesen. Nuovo Cimento 60 (1969) 47.M. Tyupkin. 68 (1971) 172. Bell and R.A.
m u s t violate this s y m m e t r y : p a r i t y is " s p o n t a n e o u s l y b r o k e n . M] f r o m / e f f [ A ] . 7 T h e " a n o m a l y " in odd dimensions appears as a paritynonconserving topological t e r m in the g r o u n d . Redlich Center for Theoretical Physics. we begin with the functional i n t e g r a l Z = Jd^di dA exp{ij\trF2/2 + i>(jtf + A M d 3 x ] (1) © 1983 The American Physical Society . To show that f e r m i o n s induce a g a u g e . I e s t a b lish h e r e . due to lef^A. ip]. l[A. The a r g u m e n t goes a s follows: While the action. ±)7W'[A] — 18 W[A] i s the C h e r n . We t h e r e f o r e c o m p l e t e the p r o g r a m begun by g e n e r a l i z i n g the axial anomaly to h i g h e r even d i m e n s i o n s 6 by e s t a b l i s h i n g t h e e x i s t e n c e of a s i m i l a r p h e n o m e non in odd d i m e n s i o n s . but d o e s not m a i n t a i n gauge i n v a r i a n c e : / e ff[/l] is found to change by a n odd m u l t i ple of ir u n d e r a homotopically n o n t r i v i a l gauge t r a n s f o r m a t i o n . however. The second way i n t r o d u c e s a heavy P a u l i . U„. (ltls*slelf[A] lim^. introduces parity nonconservation in the form of a paritynonconserving.15. I find that l i m ^ ^ „ / e f f [ A .\ = ±i. we r e q u i r e e x p z / e f f [ A j to be gauge i n v a r i a n t . Massachusetts 02139 (Received 30 September 1983) The effective gauge field action due to an odd number of fermion species in threedimensional SV(N) gauge theories is shown to change by ±7TMJ under a homotopically nontrivial gauge transformation with winding number n . for an odd n u m b e r of m a s s l e s s f e r m i o n s coupled to SU(N) gauge fields in odd d i m e n s i o n s is i n v a r i a n t u n d e r s p a c e ..d i m e n s i o n a l m a s s t e r m Mtpip v i o l a t e s p a r i t y c o n s e r v a t i o n and for M±0 the f e r m i o n s have p a r i t y . b u t introducing p a r i t y n o n c o n s e r v a t i o n through Ief{[A. In both c a s e s .n o n i n v a r i a n t t e r m in the action. In a functional i n t e g r a l f o r m u l a t i o n of the t h e o r y . N. p a r i t y c o n s e r v a t i o n need not be violated. the effective action. / e f f [ ^ ] > may not be—/ e ff[A] * s obtained by i n t e g r a t i n g out f e r m i o n i c d e g r e e s of f r e e d o m .Tk. with winding n u m b e r n. thus canceling the gauge n o n i n v a r i a n c e i n / e f f [ A ] . Laboratory for Nuclear Science and Department of Physics.M\./efM. To d e m o n s t r a t e t h a t p a r i t y c o n s e r v a t i o n m u s t be violated in this t h e o r y .j\. I m a k e u s e of an i n t r i g u i n g connection b e t w e e n p a r i t y n o n c o n s e r v a tion and the n o n i n v a r i a n c e of t h e action u n d e r homotopically nontrivial gauge t r a n s f o r m a t i o n s .30. the "anomaly" causes a "physical" groundstate curr e n t to violate a s y m m e t r y of the o r i g i n a l action. which. / e f f [ A ] c h a n g e s by a n u m b e r . s o that the effective action changes by 2im u n d e r a l a r g e gauge t r a n s f o r m a tion. topological term in the effective action. The n o n c o n s e r v a t i o n of p a r i t y in odd d i m e n s i o n s i s analogous to the n o n c o n s e r v a t i o n of the a x i a l c u r r e n t in two and four d i m e n s i o n s w h e r e PauliVillars regularization introduces a mass which v i o l a t e s a x i a l s y m m e t r y . ( J / ) .V i l l a r s r e g u l a t o r field and s u b t r a c t s lirnu^„l. (Jap) contains a topological t e r m . Gauge invariance can be restored by use of PauliVillars regularization. M] — a t h r e e . M] c o n t a i n s a p a r i t y . Cambridge.. one may s i m p l y add the topological t e r m ±wW[A] to the gauge field action. which v i o l a t e s p a r i t y c o n s e r v a t i o n .Er The action. In this c a s e . although they can be e a s i l y g e n e r a l i z e d to higher odd d i m e n s i o n s .u[A. PACS numbers: 11. e . 6 — w h i c h c h a n g e s by ±vn u n d e r a h o m o topically n o n t r i v i a l gauge t r a n s f o r m a t i o n . r a t h e r than a s a topological t e r m (~*FF) in Sll(Ja>1) ( t h e r e e x i s t s no axial c u r r e n t in odd d i m e n s i o n s . 1[A. Another way to r e s t o r e gauge i n v a r i a n c e i s to w o r k with an even n u m b e r of f e r m i o n s p e c i e s . .364 VOLUME 52.s t a t e c u r r e n t (Ju *'). " The c a l c u l a t i o n s will be limited to t h r e e d i m e n s i o n s . s i n c e an even n u m b e r of f e r m i o n s in t h r e e d i m e n s i o n s can be p a i r e d to f o r m D i r a c f e r m i o n s with p a r i t y conserving m a s s t e r m s . If u n d e r a gauge t r a n s f o r m a t i o n . The f i r s t way m a i n t a i n s p a r i t y a s a good s y m m e t r y . t/i]. Massachusetts Institute of Technology. I [A. a p h y s i c a l quantity. NUMBER 1 PHYSICAL REVIEW LETTERS 2 JANUARY 19S .t i m e r e f l e c t i o n (which we s h a l l call p a r i t y ) . As an a l t e r n a t i v e to i n t r o d u c i n g a heavy P a u l i V i l l a r s r e g u l a t o r to r e s t o r e gauge i n v a r i a n c e . Since (Jatl) is equal to 6 / e f f V o A .n o n c o n s e r v i n g topological t e r m . .s t a t e c u r r e n t . that the g r o u n d .S i m o n s s e c o n d a r y c h a r a c t e r i s tic c l a s s 2 . 11. Gauge Noninyariance and Parity Nonconservation of ThreeDimensional Fermions A. i .n o n c o n s e r v i n g s p i n equal to 2 \M/\M.l~4 I find that t h e r e a r e two w a y s to r e g u l a t e the u l t r a v i o l e t d i v e r g e n c e s in the calculation of I„a[A). the p a r a m e t e r s of the theory m u s t be q u a n tized s o that n u m b e r i s an i n t e g r a l multiple of 27. i s i n v a r i a n t u n d e r local SU(A0 gauge t r a n s f o r m a t i o n s .MJ). no y s ). however. which is not s u r p r i s i n g .
will therefore change sign if the number of eigenvalues which flow from positive to negative values (or vice versa) is odd. We now recognize that the family of vector potentials.t'Mt 5 2 . as T is varied from °° to + °°.« to the pure gauge A^x". the eigenvalues of z$ 4 may become rearranged as ~ goes from °° to +°°. T) = 0 at r = . (It is of course assumed that there is no zero eigenvalue. (3) +lcff[A])}  (2) I use the usual matrix notation A = gTaAii". where T" are antiHermltian generators of the group. b e cause U„ is not continuously deformable to the identity.4. parametrized by T. 1). defined as the product of the positive eigenvalues of 04 at T = °°.4. The spectrum of i04 [A] at T = =° is identical to the spectrum at T = + oo. and the fermions are in the fundamental r e p r e sentation.2. z'o2. and we consider gauge transformations which approach the identity at large distances. 3 We use here a Euclidean formulation of the threedimensional theory. in the gauge A4 = 0 [the space x ' = (xf'. (8) Since there exists a 4 x 4 matrix which anticommutes with z'#4. The eigenvalues of it>i are plotted along the vertical axis. T). £ = 1. we choose ^(x". Therefore. we observe that det «'ap(3(J +Ali). (6) with y =y 0 / / 0 To see this. For definiteness. Z. ioj. One initially positive eigenvalue is shown to cross zero as r goes from —•» to + °o 19 . one or more eigenvalues which are positive at T = °° may cross zero and become negative at T = +°° (see Fig. The Dirac matrices in three dimensions are Pauli matrices (a3. By (3) this is equivalent to U„:l<. the eigenvalues X(T) vary continuous T. The square root of the determinant. 8 this procedure maintains parity as a good symmetry. A'. from A v(xp. where Un belongs to the nth homotopy class {U„ has winding number n). we write the Dirac equation #<f> = 0 as dijj/dT = yTM4ip. T) vary adiabatically as a function of T =x4 along the path considered above.R 1 PHYSICAL REVIEW LETTERS 2 JAM AKY 1984 and integrate over the fermion fields: Z = JdA exp{z ( J tr(F2/2)d3x where Iefi[A]= Undettf +4). Av. More precisely. The remaining components of A'(x M. we work with SU(2) and a doublet of fermions—where Ta=o"/2i. the effective action Iefi_A] are not gauge invariant. 1.!![A]~Itff[A]±Tt\n\. we follow closely the analogous calculations performed by Witten in four dimensions. detl/2i$[A]. 1 = 1. Nl'MHF.Vol. may be written det^z'i^. T) =/{T)IPT(X>1). U„. Therefore. (7) Equation (7) is soluble in the adiabatic approximation.C D r = + to FIG. To begin. with winding number n. and o° are the Pauli matrices—but the results hold in any group for which n 3 is the additive group of integers. We now vary the gauge field along a continuous path.3.) Since detz'#4 can be regulated by the introduction of a parityinvariant Dirac mass. the gauge field must pass through configura Since the spectrum of y T $ 4 is equal to the spectrum of z'$4. 3. T). hence the base manifold is S3 rather than R3. is equivalent to an instantonlike fourdimensional gauge field. for a particular gauge field. n = 1. where $ 4 = 7p( 8 p +A p ) and 0 (5) tions in field space which are not pure gauge. 2.3. I now demonstrate that det (0+4) and. is the cylinder S 3 x/j]. we may define the square root as the product of the positive eigenvalues of iJ04[A^]. I show that det(0+4)(l)l"ldet(0+4O (4) under a homotopically nontrivial gauge transformation. the spectrum of z'$4 is symmetric about zero (and is real because is iUt is Hermitian). However. T) = U„1dlIU„ at r = + °°. In particular.2. by (3). where ^ T (x p ) satisfies the eigenvalue equation yT^T(xi')=X(T)cpr(xn. To prove (4). Av{xv. The number of zero crossings of the eigenvalues of 0^A\T)] is related to the number of normalizable zero modes of the fourdimensional operator # = y f ( 9 f +A().
NUMBKR 1 PHYSICAL REVIEW LETTERS 2 JANIARV 19X4 ly along the curves of Fig. where n is the instanton number (the winding number of U„).10 . existence of the nonanalytic expression. one may calculate lim^. The calculational procedure is the threedimensional analog of the EulerHeisenberg calculation of the effective action for constant field strength. and these two graphs produce ±7rW'[/lj. the paritynonconserving term. The number of zero modes of #[A'} is well known from instanton studies. does not vanish. in the special case of gauge fields with constant field strength tensor. constant field strength. First. Eq. the total number of zero modes. M] in perturbation theory 2. (10) (9) Second. I have performed two types of calculations—details will be given e l s e where. 1 2 For the even more special case of an Abelian.„ Ieft[A. it is necessary to introduce a paritynonconserving fermion mass. y5(p± = ±ip±. NT. that gaugeinvariant regularization procedures. parity nonconservation occurs in < J*1) in QED as well. Since only the vacuum polarization graph and the triangle graph are divergent. First. the effective theory in three dimensions. is characteristic of threedimensional gauge theories coupled to massless fermions. but it violates parity conservation explicitly since *FP is a pseudovector. 2 When the mass is set equal to zero at the end of the calculation. In performing this calculation. 12 At high temperatures.366 VOLUME 52. fourdimensional field theories effectively reduce to threedimensional theories. The relativistic Dirac equation used here appears also in a nonrelativistic system as is seen in the discovery of crystals which model the fourdimensional anomaly. M ] = ±nW[A]+I'[A]. R where /NA[A] parity conserving. 2 ' 13 Not only must (13) where / e f f is finite and W[A] is the paritynonconserving ChernSimons term W[A] =(l/87r2) j cfx tr(*F„ A" jA»AvAaepva). M.solution normalizable. such as PauliVillars regularization.1(T)) which pass from negative to positive (or from positive to negative) values as T is varied from °° to +°°. 20 t ja . I find 'eff*= ' e f f W . To show. . but is nonanalytic in the gauge field. restore gauge invariance at the cost of introducing parity nonconservation.» is this. (12) which is odd if the winding number n is odd. then ••±W[A]+ISA[A] is (15) (ID n_—n+=n. to regulate divergences in a gaugeinvariant manner. f " . Therefore. Therefore. I conclude that /' [A] in (13) must be parity conserving. however. This completes the proof of (4): The determinant changes sign under a large gauge transformation with odd winding number. ±7rW[A].) The K[A\. where such terms are known to appear in partial sums of infrareddivergent Feynman diagrams. (14) with *F'' = 5€ " F v a . The induced topological term ±W[A] in Ief. the violation of spacetime reflection in three dimensions may have direct. 9 If we choose the eigenfunctions of $ip± = 0 to be eigenfunctions of y5.[A] is known to produce a mass for the gauge fields. 11 I find 'eff Only if A is positive for T = + °° and negative for T = . there exists a one to one c o r r e spondence between the number of normalizable zero modes of ^[^4'J and the number of eigenvalues of i$i[A. of $ is NT = 2nt+n. (The sign depends upon the sign of the regulator mass.U m / e f f [ A . where models of vortexparticle interactions in superconductors are mathematically equivalent to threedimensional QED—while the discussion here is limited to SU(A0 theories. obtained by "integrating out" the fermions. The discovery of anomalous parity nonconservation in three dimensions has wide ranging consequences. the effective action (3) can be calculated exactly. 1. as presented by Schwinger. In the adiabatic approximation. Although fermions are known to decouple at high temperatures. measurable consequences in condensed matter physics. (7) becomes d//rfr=A(r)/(T) which has the solution f(T)=f{0)exp[£TdTX(T')]. and denote the number of zero modes of ip+ (^_) byre+ (n_). may still be relevant in the study of hightemperature fourdimensional theories. the groundstate current is given exactly by (16) <Ja) = (g>/Bv)*Fc This current is identically conserved.
R e v . E. Lett. Schonfeld. B 1 8 5 . H a r v e y . P h y s . K e p h a r t . W i t t e n . J a c k i w . P h y s . may be possible without the need for a condensate of fermion pairs. . Witten (to be published). 8 2 . Affleck. 6 P . H. J . Rev. I. K a m a l ( P l e n u m . 1 0 3 B . 7 I would like to thank R. Jackiw and C. M a s s a c h u s e t t s I n s t i t u t e of Technology R e p o r t No. (N. C. F r a m p t o n and T. B223. 324 (1982). 357 (1983). 5 1 . P h y s . B206. M. W. T e m p l e t o n . . 4 E . 1983). 372 (1982). T e m p l e t o n . This work was supported in part through funds provided by the U. Lett. 664 (1951). D ' H o k e r and E . C a m b r i d g e . Niemi and Semenoff have shown how to derive the results presented here using the anomaly in two dimensions. S. 10 I. D'Hoker and E. " J . D £ 7 . and Ann. G u e n d e l m a n and Z. Rev. T e m p l e t o n . P h y s . Semenoff for helpful conversations. 2 21 . 1343 (1983). the Meissner effect. Jackiw for suggesting this problem and for his continued assistance. W. in Particles and Fields. F a r h i . 9 F o r a r e v i e w . 1978). Department of Energy under Contract No. S. B156. C T P 1111 (to be p u b l i s h e d ) . P h y s . . J a c k i w . 1 1 7 B . Guth. as well as A. Recently. and R. Above all. and E . R. I am also grateful to E. P h y s . W. ' T h e D i r a c m o n o p o l e is the o l d e s t e x a m p l e of t h i s . Rev. P h y s . Results similar to those presented in this paper have been obtained independently by L. N u c l . Nohl. 3 7 . 422 (1983). AlvarezGaume' and E. P h y s . Farhi have shown how to compensate for the gauge noninvariance of the fermion determinant in fourdimensional theories by the addition of bosonic terms to the Lagrangian. 975 (1982). and P h y s . Lett. Rev. P h y s . Rev. s e e R.2 S. E . W i t t e n . 157 (1981). D 24. R a d u l o v i c . 413 (1982). R. Semenoff. 132 (1981). Lett. D 2j3. J . 3 E. W. the discovery of anomalous parity nonconservation in three dimensions provides a different and simpler laboratory in which to study the subtle interplay between "anomalous" violation of symmetries and global topological properties of gauge theories. S c h w i n g e r . J a c k i w . 3134 (1981). 135 (1979). Lett. H. 5 R. this implies that the analogous phenomenon in solid state physics. 8 T h i s a r g u m e n t i s due to E . R e b b i . Lett. W i t t e n . W a n . P h y s . Since the gauge fields acquire a mass without the need for Higgs fields. Guth and G. R e v . New Y o r k . P h y s . 5 0 . J a c k i w and S. R e b b i . S i e g e l . and emphasizing the necessity of parity nonconservation. 48. N i e m i and G. but the gauge fields must become massive as well. N u c l . Nucl. K. Jaffe ( M a s s a c h u s e t t s I n s t i t u t e of Technology P r e s s .PHYSICAL REVIEW LETTERS 2 JANUARY 1984 parity conservation be violated in odddimensional theories with an odd number of fermions. 2291 (1981). Boal and A. Rev. edited by A. P h y s . 7 A. 14 Also. P h y s . edited by D. in Asymptotic Realms of Physics. and C. 172 (1976).) 140.3 W.Y. Witten for communicating to me his results. 2077 (1983). M a s s . T. D e s e r . Nucl. W i t t e n . and S. R. P h y s . 14 E. DEAC0276ERO3069.
and result in identification with the mathematical concept of connections on principal fiber bundles. The monopole discussion leads to the recognition that in g e n e r a l the phase factor (and indeed the vector potential A J con only be p r o p e r l y defined in each of many overlapping r e g i o n s of s p a c e t i m e . in its global ramifications. In Sec. V.t i m e region. This concept has been taken 3 a s the b a s i s of the definition of a gauge field.368 PHYSICAL REVIEW D V O L U M E 1 2 . It is r e m a r k a b l e that these concepts have a l r e a d y been intensively studied a s mathematical constructs. Our e m p h a s i s i s on the nonintegrable e l e c t r o m a g n e t i c p h a s e factor (which does not depend on such quant i t i e s a s the energy of the e l e c t r o n ) . State University of New York. i s d i s c u s s e d in Sec. To extend the concept to 12 . describes e l e c t r o m a g n e t i s m . The famous BohmAharonov e x p e r i m e n t . The d i s c u s s i o n s in Ref. completely d e s c r i b e all e l e c t r o magnetic effects on the wave function of the e l e c tron. however. VI and VII. This concept. among o t h e r things. In the overlap of any two r e g i o n s t h e r e e x i s t s a gauge t r a n s f o r m a t i o n relating the p h a s e f a c t o r s defined for the two r e g i o n s . first beautifully p e r f o r m e d by C h a m b e r s . It leads to the definition of global gauges and global gauge transformations. probably implicitly r e c o g n i z e d by many authors. But the additional information i s not m e a s u r a b l e . Stony Brook. MOTIVATION AND INTRODUCTION The concept of the e l e c t r o m a g n e t i c field w a s conceived by F a r a d a y and Maxwell to d e s c r i b e electromagnetic effects in a s p a c e . This simple point. depends on the e l e c t r o n e n e r g y ) . We d e m o n s t r a t e how the quantization of the pole s t r e n g t h . An identification table of t e r m i n o l o g i e s i s given in Sec. A s u r p r i s i n g r e s u l t i s that the monopole types a r e quite different for SU2 and S0 3 gauge fields and for electromagnetism. Section VII d i s c u s s e s a "gedanken" generalized BohmAharonov e x p e r i m e n t for SU2 gauge fields. The demons t r a t i o n i s closely r e l a t e d to that in the original D i r a c p a p e r . We should e m p h a s i z e that our int e r e s t in this p a p e r does not lie in the beautiful. Cambridge. i s physically meaningful. N U M B E R 12 15 D E C E M B E R 1975 Concept of nonintegrable phase factors and global formulation of gauge fields Tai Tsun Wu* Gordon McKay Laboratory. In the present p a p e r we wish to d i s c u s s this question and also i t s g e n e r a l i z a t i o n to nonAbelian gauge fields. in quantum t h e o r y . D i r a c d i s c u s s e d the p h a s e factor of the wave function of an e l e c t r o n (which. T h i s r a i s e s the question of what constitutes an intrinsic and complete description of e l e c t r o m a g n e t i s m . pathdependent) Phase factor a s the b a s i s of a d e s c r i p t i o n of e l e c tromagnetism. The s p e c i a l c a s e s of SU2 and S0 3 gauge fields a r e d i s c u s s e d in S e e s . IV. This d i s c u s sion i s m a d e m o r e p r e c i s e in Sec. a striking r e s u l t due to D i r a c . An examination of the BohmAharonov e x p e r i ment indicates that in fact only the phase factor exp (^/A"d*U)' (2) and not the phase (1). deep. Generalizations to nonAbelian groups are carried out. V g e n e r a l i z a t i o n s to nonAbelian gauge groups a r e m a d e . is studied through an examination of Dirac's magnetic monopole field. R a t h e r we a r e concerned with the n e c e s s a r y concepts to describe the physics of gauge theories.1 however. Harvard University. c e n t e r e d only on the local p r o p erties of gauge fields. I.e. the p h a s e (1) contains m o r e information than the p h a s e factor (2). 3 . the field s t r e n g h t / M l . t h a t / B 1 / by itself does not. 3845 tfA^ (1) around an unshrinkable loop. It w a s l a t e r r e a l ized. and g e n e r a l m a t h e m a t i c a l development in fiber bundle theory. According to this concept. 4 is u n d e r s t o o d in this concept of e l e c t r o m a g n e t i s m . It leads to the concept of nonintegrable (i.. IH the field produced by a magnetic monopole. Massachusetts 02138 Chen Ning Yang 1 Institute for Theoretical Physics. The m a t h e m a t i c s of these r e s u l t s i s in fact well known to the m a t h e m a t i c i a n s in fiber bundle theory. 2 showed that in a multiply connected region w h e r e fuv = 0 everywhere t h e r e a r e physical e x p e r i m e n t s for which the outcome depends on the loop integral global p r o b l e m s we analyze in Sec. In other w o r d s . New York 11794 (Received 8 September 1975) Through an examination of the BohmAharonov experiment an intrinsic and complete description of electromagnetism in a spacetime region is formulated in terms of a nonintegrable phase factor. II.
i . II. e . &a= 2rr x integer and S i s s i n g l e valued. We shall now show the following: e v e r y t i m e one goes around the cylinder. What p r o vides a complete d e s c r i p t i o n that i s n e i t h e r too much nor too little is the p h a s e factor (2). The s a m e a r g u m e n t obviously holds if one s t u d i e s the wave function of an interacting s y s t e m of p a r t i c l e s provided the c h a r g e s of the p a r t i c l e s a r e all i n t e g r a l multiples of e. FIELD DUE TO A MAGNETIC MONOPOLE The definition of a nonintegrable p h a s e factor (7) in a g e n e r a l c a s e may p r e s e n t p r o b l e m s . Theorem 1: If (3) i s satisfied. Thus two c a s e s a and 6 for which n n . E x p r e s s i o n (2) i s l e s s e a s y to u s e (especially when one m a k e s g e n e r a l i z a t i o n s to nonAbelian groups) a s a fundamental concept than the concept of a p h a s e factor for any path from P to Q # =eXp " fe//' & " (7) provided that an arbitrary gauge transformation s=s0S = ( s j . C a s e a and c a s e 6 outside of the cylinder a r e then g a u g e . 4 Consider a static m a g netic monopole of strength g * 0 at the origin f = 0 and take the region R of s p a c e . 3. (A) or j ^ e ' " ^ . (nto. This we shall state and prove a s follows. we shall call the phase factor (7) a nonintegrable (i. and no physically observable effects would differentiate t h e m .(Ali)a i s c u r l l e s s . 1. (£")• *{hc/e) (3) w h e r e ft is the magnetic flux in the cylinder. which changes them into the c o r r e s p o n d i n g quantities for case 6. different p h a s e s in a region may d e s c r i b e the s a m e physical situation. If (3) i s satisfied. Now {All)b. We shall develop this theme further in the next s e c tion. ^ underd e s c r i b e s e l e c t r o m a g n e t i s m .\ *6 = S . i . To i l l u s t r a t e the p r o b l e m .1 ^ . — or(AV=(A±+^ (4) He 3a dx» (5) =(A ) .0 6 = integer give the s a m e i n t e r f e r e n c e fringes in the e x p e r i m e n t .i ^ S ~ exp(£La(Q)) e x p (  l £ A. . .) (6) III. i . 2). But it i s multiplevalued with an i n c r e m e n t of ^a=Ycj[(A^(A^]dxU tic does not change the prediction of the outcome of any physical m e a s u r e m e n t s .t i m e minus the o r i gin f = 0 . the e x p e r i m e n t is not feasible u n l e s s the m a s s of the gauge p a r t i c l e v a n i s h e s .3846 TAI T S U N WU A N D C H E N NING YANG 12 Unfortunately. 1) in a doubly connected region w h e r e the e l e c t r o magnetic field i s z e r o . S m u s t be singlevalued.*) «*(£««) (8) F o r this gauge transformation to be definable. but a itself need not b e . hence (5) can always be solved for a. no e x p e r i m e n t outside of the cylinder can differentiate between c a s e s a and 6. pathdependent) p h a s e factor. We look for a gauge t r a n s f o r m a t i o n on the e l e c t r o n wave function tpa and the v e c t o r potential (A M ) a for c a s e a. e . which is equal to exp electron beam interference plane FIG.. . let u s study the magnetic monopole field of D i r a c . we t r y to f i n d S = e " i a such that We conclude: (a) The field s t r e n g t h / . As predicted 1 by Aharonov and Bohm.fr) The phase (1) o v e r d e s c r i b e s e l e c t r o m a g n e t i s m .t r a n s f o r m a b l e into each o t h e r . In the l a s t section we make s e v e r a l r e m a r k s .e. Thus we have shown the validity of Theorem 1. A magnetic flux is in the cylinder. BohmAharonov experiment (Refs. e . Following Ref. DESCRIPTION OF ELECTROMAGNETISM The BohmAharonov e x p e r i m e n t e x p l o r e s the e l e c t r o m a g n e t i c effect on an electron b e a m (Fig. C o n s i d e r f i r s t an electron outside of the cylind e r . Electromagnetism is thus the gaugeinvariant manifestation of a nonintegrable phase factor. different physical situations in a region may have the same f m. Outside of the cylinder the field strength f)lv=0. 1.t i m e u n d e r consideration to be all s p a c e . the fringe shift is dependent on the p h a s e factor (2).
8) = 2itg(\ . With (13) w e have S a t = exp(i J D0). 2. F o r a path that c r i s s c r o s s e s in and out of the overlapping region. 3847 Theorem 2: T h e r e does not exist a s i n g u i a r i t y free A .. 7 alW with an o v e r l a p extending throughout 7r/2 . = 0 .ir) = 4vg.8 which m e r e l y s t a t e s that (Au)a and <AJt a r e r e l a t e d by a gauge t r a n s f o r m a t i o n with the t r a n s f o r m a t i o n factor (12). One possible choice i s to take the r e g i o n s to be Ra: O£0<TT/2 + 6 0 < r . denoted by n ( r . It can be r e s o l v e d in the following way. fi(r.6 < 8 < r r / 2 + 6 . One only h a s to b e careful that if the s t a r t ing and terminating point A i s in the overlapping r e g i o n . (12') But at 8 = ir the loop s h r i n k s to a point. i s r e g a r d e d a s two points Pa and Pb. w h e r e fi i s the magnetic flux through a c a p b o r  . w h e r e a point in the overlapping r e g i o n .(B)*J. At 6 = 0. such a s point P. so that (i) t h e i r c u r l s a r e equal to the magnetic field and (ii) in the overlapping region (Au)a and {AJt. = *.. or $ . such a s A — B — C — D~E in F i g . e.). i . If a s i n g u l a r i t y . I n c r e a s i n g 8 leads to a continuous i n c r e a s e in Q till one a p p r o a c h e s 8 = it.= ( A J . / To define the phase factor for a path we r e f e r to F i g . in the integrand in the exponent. a t which n(r. (A A = ^ ( 1 +COS0). T h e r e f o r e Q(r. e .e. (A. With an A^ which h a s s i n g u l a r i t i e s . each s i n g u l a r i t y . we d e fine $ along the path by (7) with (A„)a o r (Ajt.cosd).0) = 0.S. P S(P) . If the path Q — P i s e n t i r e l y within the overlapping region we have then two possible phase f a c t o r s $ Q p and $ 0 _ .f r e e A B does exist throughout R. It i s easy to prove that tQ^SW*^ i. 8) f o r r>0.370 12 CONCEPT OF NONINTEGRABLE PHASE FACTORS AND. (14') (10) Rb: TT/2 .. 2.SU»*fllCfl.). factor around the loop i s then equal to FIG. If a path i s entirely within r e g i o n a orb. The phase factor so defined s a t i s f i e s the group property. or m o r e explicitly n ( r . T h i s int e g r a l . 2. T h i s difficulty must b e resolved in o r d e r to u s e a nonintegrable p h a s e factor a s a fundamental concept to d e s c r i b e e l e c t r o m a g n e t i s m .g. rsin.ir) =0 since J4 U h a s no s i n g u l a r i t y . the phase factor i s taken to be $A BA The p h a s e and not $ . We have thus r e a c h e d a contradiction and T h e o r e m 2 i s proved.. consider the loop i n t e g r a l § Audx* for time t = 0 around a c i r c l e at fixed s p h e r i c a l c o o r d i n a t e s r and 8 with a z i m u t h a l angle <p = 0 — 2n. (We a s s u m e 0 < 8 S 7 r / 2 .f r e e in their r e s p e c t i v e r e g i o n s . tic (13) (12) which i s D i r a c ' s quantization. 0 < * < 2 7 r . the definition of $ i s $EDCBA= *M. o v e r a l l R. i s equal to the magnetic flux through a cap bounded by the loop.= (A T ). Schematic diagram illustrating the relationship between Ra and Rb. (15) (ID {AX = (Ar)b = <Ae)b = 0. g (1COS0). a l l . L e t u s seek to divide R into two overlapping regions Ra and Rb and to define (A u ) a and (Au)b. the nonintegrable p h a s e factor b e c o m e s undefined if the path goes through a singularity.. ^EDCBA" = $EDa$DaCBA $EDb$DbCBA etC = $EDC$CBA>  (16) The relationship between the e l e c t r o m a g n e t i c field and the p h a s e factor around a loop i s the s a m e a s u s u a l . (14) *0aPaS(P) = S(Q)$QbPb.iil. 2g1_ integer = D. (9) The gauge t r a n s f o r m a t i o n in the o v e r l a p of the two regions i s S=S a b = e x p H a) = exp ( ^ r 4>) T h i s i s an allowed gauge t r a n s f o r m a t i o n if and only if S i s singlevalued. a r e r e l a t e d by u gauge t r a n s f o r m a t i o n . Notice that fixing the path but sliding the points B and D along it does not change $EDCBA [because of f o r m u l a s like (14')] so long a s B and D r e m a i n in the overlapping region. 0 £ <} < 2TT .a < 6 ^ 7 0 < r . . ) Take (A.
3. Theorem 3: If (13) is not satisfied (the above method of resolving the difficulty would not work since) there exists no division of R into overlapping regions Ra. provided Dirac's quantization condition (13) is satisfied. . then it is possible to have any value for g. resulting in a new po tential (Aj'f We shall illustrate schematically the transformation by "elevating" the region 6 in Figure 3(a). . Thus two different gauges may share the same characterizations (a) and (b). 8 fixed. One could also contract it. of course. which results in Fig. We have satisfactorily resolved the difficulty mentioned at the beginning of this section. . Thus we have a contradiction. Theorem 3 shows that if Dirac's quantization condition (13) is not satisfied. if one excludes the halfline xy =0. We shall now prove the following. t = 0.°>]«v\jcw™ «]• (17) This is not equal to unity when B = ir. properly generalized to the case of more than two regions. then the field of a magnetic monopole of strength g cannot be taken as a realizable physical situation in R. but not necessarily analytic).) This conclusion is the same as Dirac's. To prove this statement. We shall call the process of distortion leading from one gauge to another a global gauge transformation.2?r. the phase factor is the same whichever way one chooses the cap provided it does not pass through the point r =0 (any t). provided they are gaugetransformed into each other by (12') in the region of overlap. or any halfline starting from r = 0 leading to infinity. . we had chosen the vector potential to be given by (11). 0 = 0 . One could extend the region b. Notice that because of Dirac's quantization condition. we can attach to this gauge any (/!„)„ and (An)). observe that if such a division is possible. 3(c). The phase factor defined by the generalization of (15) is equal to exp rcw. will be assumed to be many times differentiable. Both of these questions are related to gauge transformations. Notice that the original gauge we started with was characterized by (a) specifying [in (10)] the regions [Ra and Rb] and (b) specifying the gauge transformation factor (12') in the overlap (between Ra and Rb). provided the whole R remain covered One could create a new region by considering a subregion of 6 as an additional region Rc [Figure 3(b)]. would hold. and define the gauge transformation connecting them as the identity transformation so that (A(i)c= (Ajy One can then "elevate" Rc and contract Rb. since (13) is assumed to be invalid. one could generalize (15) and arrive at a satisfactory definition of the phase factor. in fact. But. Take the loop to be a parallel on the sphere r fixed. The phase factor around a loop is then a continuous function of the loop. to round out our concept of a nonintegrable phase factor the question of the flexibility in the choice of the overlapping regions and the flexibility in the choice of A B in the regions must be faced. [ A distortion may of course lead to no changes in characterizations (a) and (b). The collection of gauges that can be globally gaugetransformed into each other will be said to i i FIG. (The resultant/„„ is. Through operations of the kind mentioned in the last three paragraphs. so that condition (i) and (ii) stated above.] In the case of the monopole field. z <0. It is a natural generalization of the usual gauge transformation. It is also a concept not tied to any specific vector potential. For each choice of gauge there is a definition of a nonintegrable phase factor for every path. IV. we arrive at a large number of possibilities each with a particular choice of overlapping regions and with a particular choice of gauge transformation from the original {Aja or {Au)b to the new A^ in each region. but viewed from a somewhat different point of emphasis. GENERAL DEFINITION OF GAUGE AND GLOBAL GAUGE TRANSFORMATION Assuming that (13) holds. extended to deal with the intricacies of the field of a magnetic monopole.371 3848 TAI TSUN WU AND CHEN NING YANG 12 dered by the loop.Rc. This definition is a natural generalization of the usual concept. It does not refer to any specific Aa. (Of course. not a monopole field in general. Distortions allowed in gauge transformation. Each of such possibilities will be called a gauge (or global gauge). which we shall call distortions.Rb. Consider a gauge transformation  in Rb (i. The group condition $ccBAa = $ccBb®BbAa i s always satisfied.) Thus a gauge is a concept not tied to any specific vector potential.
that Theorem 5a: The field strength f ^ is invariant under a global gauge transformation. as expected. Thus the total magnetic flux around the origin r = 0 is equal to (2irHc/e)D for any gauge field defined on 8 D . To prove Theorem 8 in this case one need only add three loop integrals to FIG. are not r e lated by a global gauge transformation. O^t^l (18) In an overlap between regions a and b this interpolating vector potential assumes values (AB)J" and (A u )j r) which are related by the proper gauge transformation belonging to this overlap. For a given value of D. We shall state this as a theorem and give another proof of it. such as the equator on a sphere r = 1. •(AJb]dx" ih~c f£sQ*sjdx» (21) which completes the proof. Case of three regions for Theorem 8. i. If one starts with any gauge which is of the same gauge type as SD. Theorem 8: Consider gauge 8D and define any gauge field on it. and the integral is taken around any loop around the origin r = 0 in the overlap between Ra and Rb.RC). .. Notice that if there are more regions in a gauge than two. (19) should be replaced by a sum of line integrals along paths that are in the various overlaps between the regions. the gauge defined by (10) and (12) will be denoted by SD. Theorem 6: For D *D'. + (l^)(A„)« J . or rather the lack of relationship. which contradict Theorem 3. To prove Theorem 7. The total magnetic flux through a sphere around the origin r = 0 is independent of where S is the gauge transformation defined by (12) for the gauge SD in question. To prove this theorem we use Theorem 7. For a case of three regions there are three paths. Along each path the integral is of the form (19) with S denoting the gauge transformation factor. Thus (19). Using Theorem 7 we interpolate between them and obtain unquantized magnetic monopoles. for Abelian gauge fields. and makes a global gauge transformation on it.i? 4 ). Now go back to Theorem 6 and assume it to be invalid.372 C O N C E P T OF NONINTEGRABLE PHASE FACTORS AND. we simply make a linear interpolaticT between the two original gauge fields which we shall denote by (AM)(co and (AM)(6): A"' . and (Rc . Then we can gaugetransform the vector potential belonging to the monopole of strength D'hc/2e to the gauge %D. (R„. the righthand side of (21) is equal to 4itg. the gauge field and only depends on the gauge: $ f^dx*dx"==i^ j ^(\nSJdx\ (19) 12 3849 belong to the same gauge type.Ra). in addition to possible fields produced by electric charges and currents. For D*D'. Theorem 7: Between two gauge fields defined on the same gauge there exists a continuous interpolating gauge field defined on the same gauge. (20a) The flux through the lower half is equal to a similar integral around the equator: <Aa)bdx» • / (20b) Hence total flux '/[<*„). 8D and 8D. which are illustrated in Fig. by taking an infinitesimal loop.e. between SD and %D. The phase factor around a loop starts and ends at the same point in the same region. 4. For this gauge we have then two monopole fields of different pole strengths. which depends only on the gauge. we have. the followingTheorem 4a: The phase factor around any loop is invariant under a global gauge transformation. they are not of the same gauge type. the theorem itself does not refer to any specific gauge fields at all. Using (13) and (12). By the same argument as used in the proof of Theorem 7. To prove this theorem we observe that the flux through the upper half of the sphere r = 1 is equal to the following integral around the equator: / (Alife*. such as (12).(A 11 ) (a. is shown by Theorem 6. = . between the two regions containing the path. 4. is in fact the same for all gauges of the same type. i e . . the relationship. The three paths from P to Q are in the three overlapping regions between (iJ a . Thus we have proved Theorem 7. the total flux is not changed by Theorem 5a. any gauge field defined on 8D must have a magnetic monopole of strength DHc/2e at the excluded point r = 0. It follows trivially from this. Notice that although in this proof of Theorem 6 we have used two specific gauge fields. Thus it does n ot change under any global gauge transformation.
denoted in general by 6* . c in a RnRhDRr. the factors 6* Xk are ordered along the path from P to Q with the factor 6*(P)XS at the right end of the product. For a loop ABCA. s s c so. A translation of terminology is given in Table I. On the basis of the discussions in the preceding section. the phase factor around any loop remains in the same class. very much as in (21). The nonintegrable phase factor for a given path [The subscript "ordered" means that. each of the form of (20a) and (20b). (25) (24) 1 .. then there are gauge transformations Sab. CASE OF SU2 GAUGE FIELD For the SU2 case we take the infinitesimal generators Xk to satisfy X1X2X2X1=X3. Theorem 4: Under a global gauge transformation. The first proof we gave above of Theorem 8 is easy and is "obvious" to a physicist. The second proof is more involved but is more intrinsic. which is very deep. GENERALIZATION TO NONABELIAN GAUGE FIELD is now an element of the gauge group. The class does not depend on which point is taken as the starting point around the loop.Xh. as a generalization of (7). in the definition of the exponential in terms of a power series. both the concept of a gauge and the concept of a global gauge transformation are not tied to any specific gauge potentials. (i) The righthand side of (19) is independent of the gauge field. (22) and define the phase factor. 3.Sba. The theorem is a special case of the ChernWell theorem which evolved from the famous GaussBonnetAUendoerferWeilChern theorem. These concepts have been extensively studied by the mathematicians in the framework of more general 6 mathematical constructs. or equivalently S0 2 . Since S is singlevalued.e. A gauge is defined by (a) a particular choice of overlapping regions and (b) a particular choice of singlevalued gauge transformations Sab in the overlapping regions. $CABC=$CA'i>ABCA$AC So far we have only considered electromagnetism and described it in terms of an Abelian gauge field that corresponds to the group U1. Rb. etc.Sac. under a gauge transformation 3 1 = $ABCA~ &ABCA Z<A)$ABCA r ^). We shall still call it a phase factor. Since these phase factors do not in general commute with each other. we make the replacement ieAll~ebkllXk. Theorems 4a and 5a for the Abelian case need to be modified as follows.Sca. For a local region this has been done in Ref.] The algebraic operators Xk can be thought of as the collection of all irreducible representations of (22). and notice that along each path the integrand is always the difference of the vector potential A tf between two regions. the generalization to the nonAbelian case can be carried out without much difficulty. (1) In the overlapping region Ra(~)Rb. Theorem 5: The field strength/*^ is covariant under a global gauge transformation. It is a gaugeinvariant concept. The choice of gauge transformations clearly must satisfy the following two conditions. Extension to global considerations is our present focus of interest. and only depends on the gauge type. V. etc. by7 *a MX°i76'^)l (23) i. the gauge transformations Sba from a to 6 and Sab from b to a are related by where 1 is the identity element of the gauge group. Theorem 4 defines the class of a loop. (ii) The righthand side of (19) has as integrand the gradient of InS. The eigenvalue of iXk with the . (2) If three regions Ra.373 3850 TAI TSUN WU AND CHEN NING YANG 12 gether. s o t h a t ^ab^bc^ca' Hence changing the starting point does not change the class. a seminal development in contemporary mathematics. the integral must be equal to an integral multiple of a constant (in this case 2m). This concept is the generalization of the phase factor for electromagnetism around a loop with the magnetic flux as the exponent. VI. and Rc overlap. A r e markable fact is that these consequences remain valid in the general mathematical theorem. o As in the case of electromagnetism. 5 We want to emphasize two consequences of the theorem. Only theorem 4 is not immediately transparent. around the same loop if we change the starting point from A to C. or Aibk.
e.e)(X1sincl>X2cos<l>)]exp{TrX2). This is the generalization (see Ref. e . minimum absolute value is ±  . and Rh. Translation of terminology. This i s s i n g l e .2 0 X 3 ) . i . on the manifold of SU2 when <p v a r i e s from 0— 277. (One only has to verify this s t a t e m e n t a t 6 = 0. which i s easily done. Gauge field terminology gauge (or global gauge) gauge type gauge potential A* S 6 a (see Sec. 3851 TABLE I. 3) of the concept of electric charges and currents. and e. R. we define b)=b* = 0. Theorem 9: F o r the SU2 gauge group. The different monopole fields a r e t h e r e f o r e of the s a m e type. In the overlapping region... (28) We shall now t r y to define a D i r a c monopole field as a special SU2 field along only one i s o s p i n d i r e c tion k = 3 . A„ (29) which follows from the e x i s t e n c e of h a l f . T h u s . . wound D t i m e s . We shall choose ^ = exp[e(X1sui(l}X2cos(p)]. V) phase factor $QP field strength/*i. electric source. We shall only exhibit the global t r a n s f o r m a t i o n for the c a s e S. the gauges 9 D for different D can be t r a n s f o r m e d into each other by global gauge t r a n s f o r m a t i o n s . by a global gauge t r a n s f o r m a t i o n S may be changed to S' = 1. The phase factor (30) d e s c r i b e s a g r e a t c i r c l e . £ and rj a r e t h e r e f o r e allowed gauge t r a n s f o r m a t i o n s in. Thus " c h a r g e " of gauge p a r t i c l e minimum "charge" 2 for SU 2 . (27) The p a r t i c l e of the gauge field belongs to the a d joint r e p r e s e n t a t i o n .374 12 CONCEPT OF NONINTEGRABLE PHASE FACTORS AND. in c o n t r a s t with the situation for e l e c t r o m a g n e t i s m . The gauge potential 6* i s then defined everywhere in R a s a single r e g i o n . (31) It is easy to s e e that  is analytic in the c o o r d i n a t e s x11 at all points in Ra. He g (32) (33) where Au i s given in the two r e g i o n s (10) by (11).v a l u e d if and only if the quantization condition eg •— = integer =D nc is satisfied b e c a u s e for SU2 exp(47rXj) = 1 . Thus we have the following t h e o r e m . source a </* electro magnetism isotopic spin gauge field Dirac' s monopole quantization electromagnetism without monopole electromagnetism with monopole Bundle terminology principal coordinate bundle principal fiber bundle connection on a principal fiber bundle transition function parallel displacement cur