ANNALS OF PHYSICS, 141, 316-352 (1982)

On Kaluza-Klein Theory

ABDUS SALAM

International Centre for Theoretical Physics, Trieste, Italy, and

Imperial College, London, England

AND

J. STRATHDEE

International Centre for Theoretical Physics, Trieste, Italy

Received December 8, 1981

Assuming the compactitication of 4 + K-dimensional space-time implied in Kaluza-Klein-

type theories, we consider the case in which the internal manifold is a quotient space, G/H.
We develop normal mode expansions on the internal manifold and show that the conventional
gravitational plus Yang-Mills theory (realizing local G symmetry) is obtained in the leading
approximation. The higher terms in the expansions give rise to field theories of massive
particles. In particular, for the original Kaluza-Klein 4 + l-dimensional theory, the higher
excitations describe massive, charged, purely spin-2 particles. These belong to infinite dimen-
sional representations of an G( 1,2).

I. INTRODUCTION

The motivation for studying generally covariant field theories in space-time of

more than four dimensions is to obtain a geometrical interpretation of internal
quantum numbers such as electric charge, i.e., to place them in the same context as
energy and momentum [ 1, 21. The latter observables are associated with translational
symmetry in 4-dimensional Minkowski space; the internal observables would be
associated with symmetry motions in the extra dimensions.
In theories of the Kaluza-Klein type one starts with the hypothesis that spacetime
has 4 + K dimensions. One assumes general covariance and adopts the 4 + K-
dimensional curvature scalar as the Lagrangian. Next, one supposes that because of
some dynamical mechanism, the ground state of this system is partially compactifled
[3], M4 x BK rather than I’U~+~, where M4 denotes 4-dimensional Minkowski space
and BK is a compact K-dimensional space [4]. The size of B” must be sufficiently
small to render it unresolvable at the currently available energies. For example, with
the size of B” < 10-i’ cm, it would be invisible to probes of energy <lo3 GeV, this
being an upper limit.
Models have been constructed which illustrate this phenomenon of spontaneous
316
0003-49 16/82/080316-37$0500/O
Copyright 0 1982 by Academic Press, Inc.
MI rights of reproduction in any form reserved.
 ON KALUZA-KLEIN THEORY 317

cornpactilication [S, 61. In general, it is necessary for the system to include non-
geometrical (matter) fields, coupled to the metric field, for this effect to be triggered.
The presence of such ad hoc matter fields detracts from the simplicity of the purely
geometrical theory and perhaps indicates that the geometrical component is not truly
fundamental. An interesting exception, however, could be the extended supergravities
where certain matter-like fields are justified by an underlying supersymmetry [7] and
possibly the related geometry of superspace.
Assuming that the compactilication is achieved, what are the implications of taking
the extra dimensions seriously? By expanding the fields in a Kaluza-Klein theory in
terms of normal modes on the compact space BK, the extra dimensions make their
appearance as massive multiplets of the associated symmetry group. In this paper we
wish to examine these multiplets and the higher massive excitations.
To illustrate the emergence of masses, consider the original theory of Kaluza and
Klein which involves one new co-ordinate, y. The 5-dimensional line element was
assumed to take form

ds* = (dx”e,“(x))* - (dy - dx”?cA,(x))*,

where m, n, a, b ,... take the values 0, 1, 2, 3 and e,‘(x) represents the vierbein field
on 4-dimensional spacetime. The 4-vector ,4,,,(x) is to be interpreted as the elec-
tromagnetic potential. Gauge transformations here take the form of coordinate
transformations
xm --+ xm,

Y-+Y +A(xX
with

ema --+ ema,

A,-+Am + +I,

which leave the l-forms, dxmema and dy - dx%A,, invariant. The invariant coupling
to a complex scalar field, 4(x, y), would be described by the Lagrangian

If one suppose that the y-dependence of qi is such that

then this Lagrangian reduces to that for a charged scalar,

dete [le,“(ad - ieA,#)I’ - (P’ +-$I 141~1

T
318 SALAM AND STRATHDEE

where the charge is given by

K
C?=-.
R

The electric charge is hereby understood in terms of the radius, R, of the extra
dimension, interpreted as a circle. Conversely, given e, this equation fixes R (the size
of the manifold) as rc/e where K is Planck length. This also implies that the masses
are in Planck units -IC-’ - 10” GeV. Because of this one is naturally interested to
discover any exceptional cases where, for reasons of symmetry, the mass contribution
is suppressed. An example is the vector A, itself which describes a massless particle,
the photon. The gauge symmetry here associated with the transformation
y -+ y + A (x) prevents it from acquiring any mass.
To consider non-Abelian symmetries one has to interpret the new co-ordinates y”,
p = 1, 2,..., K as a parametrization of the manifold of a compact non-Abelian group,
G. The line element is now assumed to take the form [4]

ds2 = (dx”ema(x))’ - ((dy’ - dx”~A,“(x) K,“(y)) e,O(y))‘,

where the auxiliary functions e,4(y) constitute an orthonormal K-bein on the group
manifold G, and the functions K,“(y) are Killing vectors on G. This formalism will
be reviewed in some detail in the following sections.
A further generalization is possible where the new co-ordinates are taken to
parametrize some homogeneous space, G/H, on which G can act [8,9]. A novel
aspect of this kind of scheme concerns the problem of embedding. In order to give a
G-invariant meaning to the ground state it is necessary to associate the motions of
G/H with frame rotations (or tangent space transformations). This requires that the
stability group, H, be embedded in the tangent space group, SO(K), in such a way
that the K-vector of SO(K) has the H-content of G/H.
In all these theories the metric field carries an infinite number of new degrees of
freedom corresponding to the propagation of excitations in the new dimensions. As a
rule these excitations are ignored because of their large masses. However, the massive
excitations are not always ignorable: their contributions to a typical scattering
amplitude are comparable to those due to graviton exchange. To see this, consider the
scattering of charged scalar particles with the interactions represented schematically
by
%t ~Kh,,,,,~,,,~*~,~ + ieA,,,(*~,,,# + a..,

where h,, denotes the graviton field and A,,, one of the massive excitations. The
amplitude for graviton exchange is of order K*E*, while that for A-exchange is of
order e*E*/M*, where E represents the energy scale of the scattering process. Since
M-e/K these contributions are of the same order.’ Only in the region of low
momentum transfer, t + E*, does the graviton amplitude, -K*E~/c, become dominant.
’ This comparability of contributions may have relevance to the problem of renormalizability of
Kaluza-Klein theories.
 ON KALUZA-KLEIN THEORY 319

The purpose of this study is to deal with some of the general features of Kaluza-
Klein theories. These include the geometry of homogeneous spaces, the transfor-
mations of fields defined over them, the structure of normal mode expansions’ on
G/H, the embedding of Yang-Mills gauge transformations in the group of general co-
ordinate transformations, and the extraction of an effective 4-dimensional
Lagrangian. Such an effective action will be particularly useful for low energy effects
ascribable to the existence of the microscopic and as yet unresolvable structure on
space-time. Since much of the recent work in this field has been couched in the
terminology of libre bundles [lo] it is not accessible to those without expertise in
such matters. We shall therefore adopt a less sophisticated approach and express
results in the more generally familiar component notation.
The plan of the paper is as follows. Some definitions, notational conventions and a
brief description of the 4 + K-dimensional Einstein theory are given in Section 11.
There follows in Section III a discussionof symmetry expects of the compactitication
of the extra K dimensions. Harmonic expansions on G/H are introduced to describe
the excitations. An important ingredient here is the embedding of the stability group,
H, in the tangent space group, SO(K). Integration over G/H is used to reduce the
Lagrangian to effective 4-dimensional form. Some of this is illustrated in Section IV
where the low energy sector of the 4 + K-dimensional theory is shown to contain the
4-dimensional Einstein and Yang-Mills terms. Much of the pedagogical detail is
confined to the Appendices, except for Appendix V which deals with the spectrum of
4 + 1 Kaluza-Klein theory.

II. EXTENDED GRAVITY

Let the 4 + K-dimensional spacetime be parametrized, at least in some local

region, by a set of co-ordinates, 2’. On this space let there be defined at each point a
local frame of reference in the form of 4 + K linearly independent covariant vectors,
EMA.
The basic symmetries of the theory are twofold. Firstly, there must be general
covariance under the coordinate transformations, z”‘+ z’,~. Being covariant vectors,
the EMA transform according to

E,“(z)+E~(+~ E, A(z 1. (2.1)

Secondly, there is to be covariance under a group of z-dependent linear transfor-

mations among the vectors of the reference frame,

EMA + E;(z) = E,B(z) aBA(z). (2.2)

*Throughout this paper BK IS assumed to be a quotient space. One may eventually contemplate
internal manifolds (associated with symmetry breaking) which are less symmetrical.
320 SALAM AND STRATHDEE

In order to establish a pseudo-Riemannian geometry on the 4 + K-space it is usual to

assume that these transformations belong to the pseudo-orthogonal group,
SO(l, 3 + K). We shall adhere to this choice.
With frame rotations belonging to SO( 1, 3 + K) one is led to consider the frame-
independent combinations,

(2.3)

which define the covariant components of the metric tensor. By an appropriate choice
of basis, the SO(l, 3 + K) invariant tensor, qAB, can be represented by the diagonal
matrix,

rtAB= diag(1, -1, -l,..., -1). (2.4)

Indices are raised and lowered in the usual way. The contravariant quantities EAM
and gMN are defined as matrix inverses of EMA and g,,, respectively. (We employ the
alphabetic convention that letters M, N,... from the middle alphabet will be world
indices while those from the early alphabet, A, B,... shall be frame labels.)
The coupling of extended gravity to matter is conveniently illustrated by the case
of the fundamental spinor of SO(1,3 + K). This spinor, v(z), has 22+1”‘21
components where [K/2] denotes the largest integer @C/2. Under frame rotations the
components of w transform as an SO(1,3 + K) spinor,

‘Y(z) + v’(z) = S(a -‘I ‘Y(z), (2.5)

but, under general co-ordinate transformations they are scalars,

w(z) + w’(z’) = v(z). (2.6)

The partial derivatives aMw make a covariant world vector but they do not transform
homogeneously under frame rotations. The latter defect is cured by introducing the
spin connections B&z), which constitute a gauge field with respect to frame
rotations,

BM+ SB,S-’ + S&S-’ (2.7)

and a covariant vector with respect to co-ordinate transformations. The covariant

derivative is then defined by

V,~=(i?v+B,)y/. (2.8)

The spin connections belong to the infinitesimal algebra of SO(l, 3 + K). In the
spinor representation one can write

(2.9)
 ON KALUZA-KLEIN THEORY 321

where # is a generalized Dirac matrix, i.e.,

IY”, y” 1= 2vAB. (2.101

With these fields it is possible to construct a suitable Lagrangian density, one which
is invariant under frame rotations and a scalar density under co-ordinate transfor-
mations

pti = t det(E) 1+7/y‘~E~~V,~v + h.c. (2.11)

This much is nothing but a straightforward generalization to 4 + K dimensions of

the well-known 4-dimensional structures. However, when one comes to look for the
Yang-Mills potentials among the components of E and B, one will find some less
familiar things. For example, the “minimal coupling” of the vectors is contained
partly in the term I,?~~E,‘~L?,,,~ and partly in y/yAE,,MB,t,y. The latter term may also
contain a non-minimal, Pauli type of coupling.
Out of E and B and their first derivatives it is possible to construct two distinct
covariant objects, or “field strengths.” These are the torsion,

T ,,A,‘1= &EN.’ - &&,“.4 + E,‘B,,< A - E,\,CB,,,“, (2.12)

and the curvature.

RMN,AB,
= %fBNIABl- 3VBM,.4B, - BWIAC,BN,Cll, + Bh.,.4(.,B&BI (2.13)

(where early alphabet labels are raised, lowered and contracted with 71,~~defined in
(2.4)).
The curvature scalar (density) is given by

det(E) R = det(E) E,4”E,dyR .,,.N,.4R, (2.14)

and can serve as a Lagrangian density for the extended gravity fields. The fields E
and B may be varied independently or, alternatively, one may impose the torsion
constraints, T,wNA = 0, and eliminate B as an independent variable. The constraints
T = 0 are algebraic in B and can easily be solved. One finds

B M[BC] = %fA(Q[AB,C - Q,fK,.r + fi,CA ,B)’ (2.15)

where

QnlABlc= E,4~ME,~‘(L&,E,~~~- i),.E,t,,). (2.16)

322 SALAM AND STRATHDEE

III. COMPACTIFICATION

Having set up an action principle in 4 + K dimensions one might inquire into the
nature of the solutions to the resulting equations of motion. In a purely geometrical
theory one would expect the ground state, or vacuum, to exhibit maximal symmetry.
Indeed, one should find the 4 + K-dimensional Minkowski space. However, in the
presence of matter fields-scalars and vectors- the situation is more complicated
and it is possible to imagine that the ground state will manifest some lower
symmetry. In particular, if the theory is supersymmetric the fields which accompany
E and B may trigger a compactification of some of the dimensions in the ground
state. We shall assume that this happens and that the ground state geometry
factorizes into the product of 4-dimensional Minkowski space with a compact
homogeneous space, G/H, of K dmensions. This space is invariant under the action
of some group G which will be interpreted as an “internal” symmetry. We shall
further assume that the compact space is small enough (size -K) to justify, for most
purposes, the neglect of any excitations associated with it (i.e., that such excitations
would be too massive to be accessible at present energies). As mentioned in the
Introduction the excitations may not always be negligible. Their contributions are
generally comparable to those of gravitons and they would certainly be important if
the gravitons are quantized. The aim at this stage, however, is to develop an effective
4-dimensional theory for the long range degrees of freedom.
One way to obtain the 4-dimensional theory is to expand all field variables in
complete sets of harmonics on the homogeneous space G/H. The expansion coef-
ficients are 4-dimensional fields belonging to a sequence of representations of G.
Their dynamics is governed by a 4-dimensional action functional obtained by
integrating out the G/H-dependence and using the orthogonality properties of
harmonics on G/H. The resulting masses and couplings are necessarily G-invariant.
The scale of the masses is determined by the size of the compact manifold, and we
shall assume that all non-vanishing masses are therefore large. One would like to be
able to extract the zero modes and discard the rest. We shall consider some aspects of
this problem in the following but do not obtain a general solution. We now develop
the necessary formalism.
To distinguish the compact dimensions from the remaining four of ordinary
spacetime it is convenient to adopt some notational conventions. For the co-ordinates
write
ZM = (x”, y”) (3.1)
with middle alphabet Latin suffices, m, n,... taking the values 0, 1, 2, 3 and their
Greek counterparts, p, v,... taking the values 1, 2,... K. The compact manifold, G/H, is
parametrized by y“. The early alphabet frame labels are likewise divided into lower
case Latin and Greek.
To implement the parametrization of G/H there are available standard group-
theoretical techniques [ 111, one version of which we outline briefly. Suppose the co-
ordinates, y”, label the cosets of G with respect to the subgroup, H. That is, from
 ON KALUZA-KLEIN THEORY 323

each coset let there be chosen a representative element, L,. Multiplication from the
left by an arbitrary element g E G will generally carry L, into another coset, one for
which the representative element is L,,. This defines the so-called left translation

gL,=L,,k (3.2)

where h is an element of H. Both JJ’~ and h are determined by this equation as

function of y” and g. The explicit form of these functions depends of course, on the
choice of elements L, which represent the cosets. Such details are not usually
important and, here, we shall be able to manage with some general properties of the
parametrization. Among these, the most important is the notion of a covariant basis
on G/H.
To define a covariant basis consider the l-form, or differential,

e(y) = L,-‘dL,. (3.3)

This object belongs to the infinitesimal algebra of G and therefore can be expressed
as a linear combination of the generators, Qd,

e(y) = e”(y) QG
= We,“(y) QG. (3.4)

The generators Qa fall in two categories: the set Q, which generates the subgroup, H,
and the remainder, Q,, a = 1. 2 ,..., K, associated with the cosets, G/H. Correspon-
dingly, one writes

e(v) = e*(y) Q, + e”(v) QG. (3.5)

Of special interest are the coefficients e,“(y) which constitute a nonsingular K x K
matrix. The columns of this matrix provide the covariant components of the K
linearly independent vectors of the covariant basis. The contravariant components of
the reciprocal basis, em”, are defined by inverting the matrix ePa.
The behaviour of e(y) under left translations can be deduced from (3.2) and (3.3).
Thus, quite generally,

e(y)+ e(y’) = hL,-‘g-‘d(gL,h-‘)

= he(y) h-’ + hdh-’ + hL,-‘g-‘dgL,.h-‘. (3.6)
On substituting the expansion in terms of generators and introducing the matrices,
D,&. of the adjoint representation of G, defined by

g-'Qag=W?d Qs (3.7)
the formula (3.6) gives

e”i(y’)=eb(y)Dgd(h-‘)+(hdh-‘)” +(g-‘dg)6D~“(L,hm’). (3.8)

324 SALAM AND STRATHDEE

The components, (hdh-I)” and (g-‘dg)“, of elements in the algebra are naturally
defined by

g-‘dg = (g-‘dg)” Q6,

hdh-’ = (hdh-I)” Qs = (hdh-‘)” Q,. (3.9)
Since hdh - ’ belongs to the algebra of H, we must have (hdh -i)a = 0.
(A more detailed discussion of the properties of the “boost” elements, L,, together
with some illustrations is given in Appendix I. Curvature and torsion formulae for
G/H are derived in Appendix II, and various implications of the transformation
formula (3.8) are derived in Appendix III.)
Now consider the question of harmonic expansions on the internal space. To begin
with, if the internal space were simply the’group G itself (i.e., H = 1) it would be
natural to employ the full set of matrices of the unitary irreducible representations of
G. For the function, d(g), one would write

4(g) = c c @D,“,(g) dip. (3.10)

n P.4

where Diq is a unitary matrix of dimension d, and the sum includes all matrix
elements of all the unitary irreducible representations, g + D”(g). The coeffkients,
9”pq, are projected out by integrating over the group,

(3.11)

where & is the invariant measure normalized to volume V,.

For functions on the coset space G/H the expansion is somewhat restricted.
Typically, one is concerned with functions (i(g) that are subject to the auxiliary
symmetry
#i(k) = IDijth) #j(g), (3.12)

where h E H and ID(h) is some particular representation of H. This means that $(g),
though not strictly constant over the points in a coset, is related by a linear rule, i.e.,

9i(gl) = IDij(gl gF’> #ji(g*)

for g, and g, in the same coset. The appropriate restriction of the expansion (3.10) is
clear. It must include only those terms for which

D”(hg) = ID(h) D”(g).

For irreducible D(h) one should write

(3.13)
 ON KALUZA-KLEIN THEORY 325

where the sum includes all unitary irreducible representations of G for which Dyiq,
makes sense, i.e., all those which include D(h) on restriction to H. If D(h) is
contained more than once in a given D”(g) then a supplementary label, c[, is needed
to distinguish them. The dimension of D(h) is denoted do. The coefficients, d,“,, are
projected from the sum (3.13) by integrating over the coset space,

(3.14)

This expression can be derived from (3.11) by resolving the general element g- ’ into
the product of appropriate elements L, and h. Such a decomposition is unique once
the form of the representative element L, has been chosen. From (3.12) it follows that
the h-dependence factorizes,

4i(g)=4i(hm1Ly-‘)
= [D,(h - ‘) qbj(L, ‘).

Integration over the subgroup H can be carried out explicitly and leaves the result
(3.14).
Thus, for a set of fields v,(x, y) the general form of the harmonic expansion on
G/H is given by

(3.15)

Here 12labels the exitations.

The coefficients fields &(x) are obtained by integrating over G/H as in (3.14). If
the representation [D(h) to which vi belongs is reducible then it would be necessary to
separate vi into irreducible parts and make an expansion of the form (3.15) for each.
This raises the important question of how the H content of I&, y) is to be deter-
mined. It will be settled by invoking the ground state symmetry.
The full set of fields in 4 + K dimensions is supposed to be governed by an action
functional which is invariant with respect to both the general co-ordinate transfor-
mations and the frame rotations. These symmetries will not be manifest in the ground
state, however. By assumption the ground state geometry is M4 x G/H, where G/H is
a compact K-dimensional manifold. In the ground state we therefore have

(3.16)

where e,” is the covariant basis on G/H defined above. The expression (3.16) is
invariant under the product of the 4-dimensional Poincart group and the local (x-
dependent) transformations of G. The Poincari invariance is exactly as in the familiar
4-dimensional theory of general relativity. The local G-invariance is compounded of
the x-dependent left translations on G/H, y + y’(x, y) and the induced tangent space
326 SALAM AND STRATHDEE

transformations, Dab(h), with h defined by (3.2). To see this, consider the action of
the combined transformation on E,“(x, y)

where D, a is a K-dimensional orthogonal matrix. In order to have

(E:“(w)) = &,=(x4)
it is only necessary to choose (E,“) = e,” and D,” = D,“(h). This follows from (3.8)
which contains the formula

dy“‘e,“(y’) = dy“ep4(y) D,“(h-I).

Thus the embedding is fixed. With the left translation, y -+ y’, must be associated the
SO(K) transformation, Dab(h), where h E H. This implies that the K-vector of SO(K)
has the H-content of G/H. More precisely, the generators, Q,, of H can be expressed
in terms of the generators, ,?F4, of SO(K) by the formula3

$ = - &,PB, (3.17)

where the coefficients c-w0are taken from the set of structure constants c%;bof G.
The Yang-Mills group of transformations described here, which leaves the ground
state invariant also preserves the form of the expansion (3.15). Thus, with v/
transforming according to

WiCx,Y)+ Vl(x3Y’)= ‘ijCh) VjCx3Y) (3.18)

it is a simple matter to derive the corresponding rule for the expansion coefficients

vJ-4 + v;;w = q&(g) w,“sw (3.19)

They belong to the various irreducible representations of G which contribute to the

expansion (3.15).
In order to express the full 4 + K-dimensional theory in its equivalent 4-
dimensional form, one should substitute expansions like (3.15) for all of the fields in
the system. The 4-dimensional Lagrangian is then obtained by integrating the 4 + K
dimensional one over the compact manifold, G/H. Integration of products of matrices

3 For an infinitesimal transformation in H we have

D,,(h) = S,, + 6h%,+,

=6,,+0,,,

where w,s = -usa. Viewed as an infinitesimal SO(K) transformation this reads (l/2) w,,Z”s = dhTQp
which implies (3.17).
 ON KALUZA-KLEIN THEORY 327

D”(L;‘) reduces, in principle, to an algebraic problem. Although it is difficult to

obtain results without going into the details of particular cases, the most important
feature, orthogonality, can be illustrated by an example.
Consider the invariant integral

z;;,‘p ‘{ r = 1 dKy det e(y) Di.i,(L,) D~~‘,.p,(L,.-‘)V (3.20)

J
FIH
where the label, i, is understood to be summed. Since the measure, dp = d”y det e, is
invariant under left translations, one can write

z& = I dPD,“,il(Ly,)
D$,,p,(Ly;‘1

=jd~D~,i~(gL,h-‘)D~~,.,,(hL,~~‘g-
‘)
= D~4(g)Z~;,b,s,Dlt:P,(g-‘)

because the dependence on h cancels. Hence Z is an invariant tensor of G. By Schur’s

lemma it is must take the form

z;;,;‘5’ = hd,&, .
To evaluate k;;, , set n = n’, p =p’ and sum over p. One obtains

where d, and do are the dimensions of Diq and ID,, respectively. Thus

(3.21)

This result is needed for inverting the expansion (3.15).

The argument leading to (3.21) generalizes. For example, one can write

where cili2i3
is an invariant tensor of H and the 3-n symbols are essentially
Clebsch-Gordan coefficients for the unitary representations of G. The factor anlnznr
depends on normalization conventions for the 3-n symbols.

595/141/2-9
328 SALAM AND STRATHDEE

Having given, in outline,4 the general form of the expansions to be used, we shall
in the following section restrict our considerations to the leading terms. The
justification for this restriction is simply that, in the regime of low energies, only
zero-modes are important.

IV. EFFECTIVE LAGRANGIANS

It is guaranteed that the graviton and a set of Yang-Mills vectors are included
among the zero-mass fields of the 4-dimensional effective theory. This results from
the gauge symmetries of the 4 + K-dimensional theory and the assumed symmetry of
the ground state. It can be proved as. follows. For EMA(x, y) consider the ansatz [ 91

J%%Y)= (Em=(x)
() --4m%)WV,)
e,“(y) 1’ (4-l)

where the y-dependence is specified. This ansatz is compatible with a subgroup of the
4 + K-dimensional symmetries: the 4-dimensional general co-ordinate transfor-
mations,
xm --t xlrn(x)

and the Yang-Mills transformations, or x-dependent left translations

Y’-+Y’%Y)

with associated frame rotations, Dab(h),

ark=@‘) =p ax" E n"(x), (4.2a)

Y)+Fe”O(y)
aYr

e/Yy’)= 7j$ e,‘(Y)D5”(h-‘). (4.2~)

Of these rules, (4.2a) requires no comment and (4.2~) has been dealt with already in
Section III. There remains (4.2b) which implies, in view of the ansatz,

3Y”
-‘4~~(x’)Df(LyJ = --&An~(x)Dg(L,) +,euqy)) D,“(P).

4 Some illustrative examples are contained in Appendix IV.

 ON KALUZA-KLEIN THEORY 329

With the help of formulae derived in Appx. 3 this rule reduces to the form

AL(x’) =g ~~4(xw1-g~ng-‘)~ (4.3)

where A, = A,‘Q,. This is precisely the rule to be expected for a Yang-Mills

potential.
The zero torsion spin connections corresponding to the ansatz (4.1) are obtained
by substituting in the formulae (2.15), (2.16)

B o[bcl = fEromEbln am% - fEIbmEcln amEn, + &cmEaln ~~~~b~

B a[Wl = -B y[ob] = f&mEb”L,%y&),
B a[~yj = AaB(Dg”&,> - DAL,) C%9) czoy,
B a[Bcl -- 0,

B a[4y] = by + CYY) c,-qy* (4.4)

In these expressions the following notations are used:

E,amEb, ’ = EamEbn - EbmEan,

F mnci = amAnd - a,Amd -AmbA&-.&

x,“(y) = e,“(y) e,“(y), (4.5)

and the structure constants of G are defined by

lQci3Q,xl= ccgtQT.
On substituting the expressions (4.4) into the 4 + K-dimensional curvature scalar
one obtains

R 4+x = R, - $F,,“f’,,6D,y(L,> Dry + R,, (4.6)

where R, = R,,(E) is the usual 4-dimensional curvature scalar, expressed in terms of

the vierbein E,‘(x), and R, is the constant curvature of G/H (see Appendix II). On
integrating y’ the Yang-Mills Lagrangian emerges, since

&(&y det e(y) Day@,) Dsy(L,) = kd,B,

K
where k = dim(G/H)/dim(G). Thus, the above assertion that the system contains the
graviton and massless Yang-Mills fields is verified.
There may also be zero mass scalars in the system, like the Brans-Dicke scalar in
the K = 1 theory. This point is sensitive, however, to the matter field couplings which
330 SALAM AND STRATHDEE

are generally needed to force the compactification. We shall not therefore discuss the
scalar modes.
Finally, to illustrate how the ansatz described above can be combined with an
analogous one for a spinor field to give an effective 4-dimensional term, we consider
the expression

g* = + det E(x, JJ) I+T~‘E~“W,,, ty + h.c. (4.7)

with r&y) represented by the ansatz,

v(x, Y> = w, - l) w(x), (4.8)

This is a typical term from the harmonic expansion of r&y). The spinor of
SO(1,3 + K) must of course be decomposed relative to the subgroup SO(1, 3) X H;
each piece in the decomposition will yield a v(x) and the details of labelling are
ignored here. One must substitute the expressions (4.1), (4.4) and (4.8) into (4.7),

gti = G det E(x) det e(y) W(x) D(L,) y’ E,“i?, + E,“a, - f B .,BCIy”‘1
[ I

+ Y”
I
eaP% - +sYlac~
r”II
’ I+, - ‘) y/(x) + h.c. (4.9)

To simplify this expression it is necessary to eliminate the derivative 8, by means of

the formula
f3,D(Ly-‘) = -D(L,-‘) a,D(L,) II(q)
= -e,“(y) NQA WY-‘)
which follows from the definition (3.3). Multiplication by eaP then gives

eaUawW-‘) = -P<Q,> + n,“WQ,-1) WY-‘),

where rc is defined in (4.5). It turns out that the n-containing terms are cancelled by
similar terms in the spin connections. This cancellation is ensured by the embedding
of H in the tangent space group (see Appendix IV). After some tedious calculations it
is found that (4.9) reduces to

Ye = $- det E(x) det e(y) P(x) D(L,) ya E,“(x) a,,, + $ Bolbcl(x) ybc
[ I

-Aad( + &'abb(x) yaby"D,"(L,-')

I

- Y”WQ,) + $ c,qyfDY
1D(L,
-‘)v(x)
+h.c. (4.10)
 ON KALUZA-KLEIN THEORY 331

and it remains only to integrate over y. Since y’ must commute with D(L,), the result
takes the form

3 vKdetE(x) VW f’ VW 8, + f Balbcl(x)Y"'-~%)WQD-~(
[ 1
+ $ Fahg(x) yabr” - M v/(x) + h.c., (4.1 1)
I
where Ts^and M are matrices defined by

(4.12)

(4.13)

Note that the “Pauli-moment term” implied by non-zero r” is a necessary conse-

quence of our formalism, and the precise numerical expressionsfor Ts and M are the
hall-marks of the expansion on G/H.

V. CONCLUSIONS

In this study we have attempted to analyze in some detail various formal aspectsof
Kaluza-Klein theories in which the ground state geometry takes the form M” X II”.
where BK is a K-dimensional quotient space G/H. The symmetries of this ground
state are used to classify the excitations of the system, and these excitations are found
to include particles of spins zero, one and two, in general. In particular, there are a
set of four vectors belonging to the adjoint representation of G contained among the
components of the metric tensor and spin connection which serve as Yang-Mills
potentials.
Harmonic expansions on BK are used to define sequencesof component fields,
defined on M4 and belonging to irreducible representations of the Yang-Mills group,
G. An important feature in determining the content of these expansions is the
embedding of the stability group H in the tangent space group SO(K). This
embedding is discussed in detail. Considerable use is made of the “boosts,” L,, in
setting up the formalism. The transformation properties of L,, are used to define the
action of G on the manifold G/H, and the derivatives of L, give the covariant basis
vectors on this manifold. They also are used in the construction of integral formulae
for mass matrices and coupling parameters for Pauli-moment-like terms in the 4-
dimensional effective theory (formulae (4.12) and (4.13)).
We remark that the ansatz for the 4-dimensional fields associated with massless
particles-graviton, Yang-Mills vectors and Brans-Dicke-type scalars-can be seen
to arise as leading terms in the harmonic expansions. The complete spectrum for the
332 SALAMAND STRATHDEE

4 + l-dimensional theory is discussed in Appendix V. It is shown there that the

massive excitations are purely of spin-2 and that they can be assigned to intinite-
dimensional representations of the non-compact group SO( 1,2). Such non-compact
symmetries are spontaneously broken and may be viewed as spectrum generating
symmetries. It is interesting that, viewed in this light, Kaluza-Klein theory seems to
constitute an example of a consistent theory of massive, charged spin-2 particles.
Information drawn from the structure of the harmonic expansions would be useful
in testing the classical stability of Kaluza-Klein type solutions. We have not dealt
with this type of investigation (except for the simple example of 4 + 1 theory) since it
is generally necessary to introduce matter fields of some sort into the system to
generate non-vanishing curvature in G/H5 Such matter fields, which must affect the
stability considerations, represent a non-geometrical element in the theory (except,
perhaps, in the case of extended supergravitities [7]) and are particular to specific
models.

APPENDIX I: NON-LINEAR REALIZATIONS

To supplement the discussion of Section III we give here a somewhat more

amplified description of the properties of non-linear realizations of continuous groups.
The concepts were thoroughly exposed in the physics literature a few years ago when
their appropriateness in treating spontaneously broken symmetries was noticed [ 121.
In that work the space on which the group acts is a set of scalar fields and the non-
linearity can be expressed through algebraic constraints on these fields. Because of
such constraints it is impossible for all components to vanish simultaneously, hence
the absence of an invariant ground state, and hence the relevance to spontaneous
symmetry breaking. From a more abstract point of view, the set of scalar fields
constitute a vector space which supports a linear representation of the symmetry. The
constraints serve to pick out a subspace, generally curved, which is carried into itself
by the action of the group. On such a subspace the group is said to be realized non-
linearly. Indeed, to characterize the fields belonging to such a curved subspace it is
generally necessary to adopt a curvilinear co-ordinate system and so to express the
group actions by the transformations which they induce in these co-ordinates.
The classic example of non-linear realizations is the treatment of chiral SU(2) x
SU(2). This group has the infinitesimal structure of SO(4) and the scalar fields (a, rc)
are assumed to transform as a 4-vector. If they are subjected to the constraint

t~~+-n~=F~, (I-1)
where F is a fixed number, then the number of independent components is reduced to
three: they belong to a 3-sphere of radius F. This 3-sphere is of course carried into

’ The original Kaluza-Klein theory in 4 + l-dimensions represents a case where G/H is flat.
Generalizations of this have been investigated by Scherk and Schwarz [ 131.
 ON KALUZA-KLEIN THEORY 333

itself by the transformations of SO(4). It is invariant. To parametrize the points of

the 3-sphere one could simply use the 3-vector x and write u = f -7. For
some purposes, however, it is more useful to define a set of boosts, L,, which are
SO(4) transformations that carry a given reference point, say

(u, n> = (F, 0) (1.2)

into a general point. in terms of the 4 x 4 orthogonal matrices I),,(g) of the vector
representation one writes
CJ= Du(L,J F,
7ci= Di,(L,) F. (1.3)

Of course there is a great deal of latitude in the choice of L, and it will generally be
necessary to use at least two co-ordinate patches if the sphere is to be covered
without generating artificial singularities.
To illustrate the parametrization of the sphere we give the example of spherical
polar co-ordinates,
a=Fcos8,
TT,= F sin 6’ cos (0,
7c2= F sin 0 sin v, cos v/,
x3 = F sin 0 sin v, sin v/, (1.4)
which corresponds to the choice of L,,
L, = ,-0Qu-rnQn,-~Q~1~
(l-5)

Here the operator Q,, = -Q,, is the generator of the infinitesimal rotations in the
@-plane

On applying an SO(4) transformation, g, to the sphere, the point whose co-

ordinates are (0, q, w) will be carried into one with co-ordinates ((7, cp’, w’). A
rotation through the angle w in the 14-plane, for example, gives

cos 8’ cos w sin w cos e

=
( sin 8’ cos cp’ 1 ( -sin w cos w )i sinBcoso ) ’
By solving these equations one can discover the transformed boost, L,,. In general
the co-ordinate transformation is given by
334 SALAM AND STRATHDEE

From this it follows that the precise relation between L, and L,, must take the form

gL, = L,sh, (1.7)

where h is an orthogonal transformation in the 123-subspace, i.e.,

D,,(h) = Kz,. (1.8)

Since L,, is known once the angles (0’, q’, w’) have been computed, the SO(3)
transformation, h = L,,-‘gL,, is well defined.
To compute the basis vectors on the 3-sphere consider the differential
e=L /dL,
=e OQolemQne*Qnd OQSe -mQne-eQu
(e- >
= -deQ,, - dp(Q12 cos 6 + Qd2 sin 6)
- dv/(Qz3 cos v, + (Q,, cos 8 - Qd3 sin 0) sin 9)
= -d$Q,, - dp sin eQ,, + day sin 0 sin (pQd3
- dqoQ,* - dv cos (pQz3 + dv cm 8 sin vQ,, . (1.9)

From this expression one extracts the coefficients

41 42 43

12 23 31

(I. 10)
0 -cosfp cos esinq

where the labelling of rows and columns is indicated. The metric tensor on the 3-
sphere is given by

ds’ = (doe,= + dqeva + dye,“)*

= de2 + sin’ t9dp2 + sin’ 0 sin* (Ddy* (I.1 1)

as would be expected in spherical polar co-ordinates. The spin connection BaleYl is

constructed out of the quantities
 ON KALUZA-KLEIN THEORY 335

= 42 (1.12)

as discussed in the text, (4.4) (and further in Appendix 2).

The case discussed above is an example of a non-linear realization in which the
group G = SO(4) is made to act on a 3sphere. The points of the 3sphere are put
into correspondence with the left cosets SO(4)/SO(3) through the parametrization of
the boosts, L,. The procedure generalizes directly to manifolds of the form G/H,
where N is some subgroup of G. (Right cosets can also be used. One writes

On the manifold of G itself one can detine the action of G X G

In all such cases the mapping y -’ y’ is determined once the form of L,. has been
specified.)
Global properties of a quotient space G/H are often made transparent by
embedding in a flat space of higher dimension, i.e., by viewing the non-linear
realization as obtained from a linear one with constraints applied. A simple
illustration of this is provided by the case of SU(3)/SU(2) x U( 1) which we consider
briefly.
The infinitesimal transformations of SU(3) are generated by the eight operators

Qa” = -Q/, a,b= 1,2,3 CQ,,’= 0). (1.13)

The boosts may be given by

L, = exp(>j”Q,-’ - h.c.), (1.14)

where the co-ordinates J+, a = 1, 2, are complex. The little group in this case,
H = SU(2) x I/( 1), is generated by Q, ‘. With this choice of L, the procedures
discussedabove could be used to find the mapping y” + y/e and the induced SU(2) x
U(1) transformation, h, corresponding to an SU(3) transformation, g. However. we
shall not pursue this.
The manifold SU(3)/SCI(2) x U(1) clearly has two complex dimensions. In order
336 SALAM AND STRATHDEE

to see that it is, in fact, CP, one can proceed as follows. Let the hermitian, traceless
matrices, YO’, be the co-ordinates of a flat space on which SU(3) acts linearly

Y+ Y’ =g Yg-1. (1.15)

An invariant subspace can be picked out by requiring the co-ordinates to satisfy the
matrix equations

where R is a constant. Since Y is hermitian, it can be diagonalized by an SU(3)

transformation and one tinds easily that its diagonalized form must be

-R/3
p= -R/3 (1.17)
i 2R/3 ) ’

In other words, the general solution of the constraints is given by

Y=L, I%- (1.18)

with appropriate SU(3) boosts, L,. On the other hand, it is possible to express the
general solution in the form

yab=f~ab-!!$)R, (1.19)
(
where qa is a complex triplet and ii” = q ,*. Since Y is unchanged by the complex
scaling q. + dqn, this means that the manifold parametrized by L, can be identified
with the 2-dimensional complex projective space, CP,.
As a supplementary remark we note that it is possible to define the action of a
non-compact group on a compact manifold. A trivial example of this is provided by
the group of displacements on the real line. By factoring out a discrete subgroup
corresponding to displacements which are integer multiples of some unit, 27~R, one
obtains a compact quotient space. Formally, one writes

L, = eyQ, 0 <y < 27rR, (1.20)

where Q is the generator of infinitesimal displacements. Applying an arbitrary

displacement, g = exp oQ, on the left, one obtains as usual,

gL, = L,,, h,
where
y’=y+w-27&z,
h = (e2nRQ)“, (1.2 1)
 ON KALUZA-KLEIN THEORY 337

and the integer n is determined such that y’ lies in the interval (0, 2nR). If the points
y = 0 and y = 27rR are identified then the quotient space is a circle of radius R.
A less trivial example is the realization of the non-compact SO(1, 2) by its action
on the projective cone, a compact l-dimensional manifold. That such a realization
should exist is easily seen by embedding the cone in 3-dimensional Minkowski space.
The null cone

Y,’ - Y,Z - Y,Z = 0 (1.22)

is clearly invariant under the action of SO(1,2). But it is a 2-dimensional space.

However, the cone is also invariant under the scaling transformation

Y,-nu,. (1.23)

A l-dimensional manifold is obtained by identifying those points which are related by

scaling. (In practice one would be dealing with functions on the cone which are
homogeneous of degree zero.) A possible parametrization of the manifold would
involve writing
1
Y,= siny . (1.24)
t cos y 1

To find the mapping y--f y’ corresponding to an hyperbolic rotation in the 02.plane,

for example, one would write

to obtain the relation

sin y
sin y’ = (1.25)
chw + she cos y .

This kind of realization can be formalized in the manner used previously. The
generators, Qab = -Qba, of SO(1,2) satisfy the commutation rules

[Qab,
Qc,l= s,cQod
- v,cQm+ vac,Qbc
- Qac. qbd (1.26)

It is possible to pick out two of these,

QOZ and Q+= Q,,+ Q,, (1.27)

338 SALAM AND STRATHDEE

which are to be identified as the generators of the (non-compact) stability group, H.

They satisfy

[QwQ+l=Q+. (1.28)
The boosts are defined by
L, = eYQ12. (1.29)
Since any element of S0(1,2) can be represented unambiguously in the form
e~Q~zeBQ+e~oz (1.30)
it will always be possible to solve the mapping relation

gL,=L,<h

for y’ and h as functions of y and g.

Another example is the flat group of Scherk and Schwarz [ 131. In its simplest
version, the generator Q, is distinguished from the others, Qj, j = 2, 3,..., K. The
commutation rules are

[QivQjI= 0,
[Qi,Q,l=MijQj, (I.3 1)

where M is antisymmetric, i.e., Q, generates orthogonal transformations while the Qi

generate translations. (For K = 3 this is simply the 2-dimensional Poincare algebra.)
To construct a compact manifold for the group to act on, consider the elements
L = ,y’Qt,yjQj
9 (1.32)
Y

where y’ ranges from 0 to 2x/m, (m, = smallest non-zero eigenvalue of iMjk) and the
y’ are restricted to the ranges

O<y’<27rR’, j = 2,..., N. (1.33)

The action of the group on L, is easily found. Firstly, the action of exp w’Ql is
given by
euLQILy = L,,, (1.34)

where y’ j = yj and y” = y’ + CIJ’ (mod 2x/m,). Secondly,

ewjQjL Y=Ly,h, (1.35)

where y” = y’ and

yfj = yj + (e-y’M)jk & - znRjnj (1.36)

 ON KALUZA-KLEIN THEORY 339

with the integers ni chosen such that y’j falls in the interval (0, 27rB’). The little
group transformation associated with this is then

h = fl (e2nRjQj)nie (1.37)

The covariant basis is defined as usual by the l-form

e = L,-‘dL,
=dY’(Q, -YjMjkQk) + dY’Qj1

I.e.,

Q, Qk
-y%li,
eun = Y’ 1 (1.38)
y’ ( 0 Ji, 1
To construct Killing vectors the matrices D, B of the adjoint representation are
needed. These are given by

D,“(L,) Q, = L,-‘QJ,
= Q, -y%kQk,
Din&y) Qa =Ly-‘Q/Ly
= <e”“>jk Qk,
I.e.,
1
D/W,> = (1.39)
0
The Killing vectors for left translations are then

L’(Y) = Da’&) e4’

1 0
(1.40)
= (
0 (exp y ‘M)jk )*
(This matrix was denoted U(y) in the work of Scherk and Schwarz.)
We close with the observation that the zero-torsion spin connection B,’ (see
Appendix 2) is given for this manifold by
B,‘=-Bjl ~0, Bjk = -dY’Mjk, (I.4 1)

and the curvature 2-form vanishes, i.e., the manifold is indeed flat.

595/141/z-IO
340 SALAM AND STRATHDEE

APPENDIX II: CURVATURE AND TORSION ON G/H

Because of their high degree of symmetry, quotient spaces have rather simple
geometrical properties and it is possible to derive a number of explicit formulae to
describe these. In this appendix we give derivations for such of these formulae as
have been referred to in the text. All of them arise out of manipulations of the boost
elements L, which we use to characterize the manifold G/H.
As has been mentioned in Section III, the first and most important step is to
construct the vectors of a covariant basis on the manifold. To this end one must
consider the l-form

e(y) = L,-‘dL,

= @i(y) Q,
= dY‘e,%) Q,, (11.1)

where the infinitesimal generators of G are denoted Qk. These generators satisfy the
commutation rules

[Qa.
3Qal= ca~‘Qp (11.2)

with structure constants c’@.

The differential properties of e(y) needed for the discussion of curvature and
torsion are summed up in the 2-form

de(y) = -dL,-’ A dL,

= L,-‘dL,L,-’ AdL Y

= e(y) A e(y), (11.3)

the Cartan-Maurer formula. On substituting the expression e = e”Q& and using the
commutation rules satisfied by the generators, one can extract the equivalent form
de& = $8 A eFc6tb (11.4)

or, in terms of components,

ape”’ -anew’ = -e,6eupcfif’. (11.5)

As explained in the text the generators Q& separate naturally into two sets, Qz
belonging to the subgroup H, and Q,, the remainder. The coefficients e,*,
a = 1, 2,..., N, constitute a set of N covariant basis vectors. The reciprocal basis is
denoted ecr”, i.e.,
eMunepv= 6,“. (11.6)
 ON KALUZA-KLEIN THEORY 341

The torsion 2-form, T”, is defined by

T” = de” + e4 A B,“, (11.7)

where the l-form B,” represents the spin connection. Once the latter quantity has
been assigned it is possible to construct, in addition to T”, the curvature 2-form R,”
defined by
R a“=dB a5+B a YAB5 Y * (Ii.8)

One of the simplest possibilities is to choose B such that the torsion vanishes,

0 = de” + e4 A B,“. (11.9)

On comparing this equation with the Cartan-Maurer formula and noting that
cDta = 0 (because H is a group) one finds immediately
B 4a=ey$ 2 yoc’+ e7cw”
= ey(4cy5a + 7ry7cMa). (II. 10)

In this last equation we have used

e? = eYnyT= eyeyee,“. (II. 11)

From the expression (11.10) for B one obtains the curvature 2-form

RDY= $ea A e’($c,,6c8qY + c,,‘cgqy + $,bsc,by). (II. 12)

(In deriving this we have used canY= 0 which is not generally true. See for example,
the discussion of SO(2, 1) in Appendix I. In the general case one would have to keep
such terms.)
Another simple possibility is to choose B such that the torsion takes the form
T” = te5 A eye5Ya . (11.13)
Comparison with the Cartan-Maurer formula this time gives
B 4a = e”c-Y5 a (11.14)
and hence the curvature

(11.15)
In particular, if the manifold is G itself, i.e., H= 1, then this assignment gives
vanishing curvature. (On the other hand, for some interesting manifolds such as
SO(N + l)/SO(N), the structure constants caDy vanish and the two assignments are
identical.)
342 SALAM AND STRATHDEE

APPENDIX III: TRANSFORMATION PROPERTIES

Some symmetry properties of the manifold G/H are needed for an understanding of
the transformation behaviour of the Yang-Mills fields. The purpose of this appendix
is to discuss the derivation of the necessary formula.
Consider firstly the transformation rule (3.8) for the l-form eb. For an x-dependent
left translation, y + y/(x, y), this reads
e”(y’) = es(y) D$+(h-I)’ + (g-*dg)6 D/(L,h-I), (III. 1)

where DGd is a matrix of the adjoint representation, i.e.,

g-'Qig=Dh9Qp
The components (g-‘dg)” and (h&-i)” are defined by

g-‘dg= (g-‘dg)“- Qa;,

hdh -I = (hdh -I)& Q6
= (hdh-‘)‘Q,, (111.2)

which is possible since g-‘dg belongs to the algebra of G while hdh -i belongs to the
subalgebra associated with H. In the following we shall assume that the algebra is
fully reducible, i.e.,
D,“(h) =0$(h) = 0. (111.3)

This means that (111.1) separates into components from G/H and H, respectively,

e”(y’) = e”(y) D,“(h-‘) + (g-ldg)d Dbn(L,h-‘), (111.4)

e”(y’) = e”(y) D,J”(~-‘) + (hdh-I)” + (g-‘dg)’ D~b(L,h-‘). (111.5)

Under the transformation y -+ y/(x, y) we have

8y’fi 8y’”
dy” = dy”- +dx”-
8Y” i3Xrn
or, equivalently,
dy" = dy'" - 3Y” +dx”-. 8Y”
tJy’” iYXrn

Partial derivatives transform such that

dy” 8, + dx”&,, = dy”’ ($&)+dx” (a,+$$,;)

(where a:, denotes differentiation with respect to x”’ at lixed y’“).

 ON KALUZA-KLEIN THEORY 343

Extracting the coefficients of d~‘~ and dx” from the formulae (111.4) and (111.5)
gives

eua(Y’) = 5 e,‘(v) D,“W’), (111.6)

0-g eLxD(y>~,“(h-‘) + (g-La,g)hD~a(Lyh~‘), (111.7)

e,“(y’) = f$ (e,“(y) DE”@ - ‘> + (ha,. h I)“), (111.8)

0=-g eL,n(y)DB”(hpl) + (ha;h--I)”

+ (g-‘iJ,,g)bf(L,h-I). (111.9)

The formula (111.7) can be arranged in the form

aYs
-=-(g-‘a,g)6Kgyv), (III. IO)
f3Xrn

where Kf, the so-called Killing vector, is defined by the expression

K6YY) = ~sy(L,) e,“(u). (III. 1 1 )

Its transformation behaviour is obtained with the help of (III.6), viz.

K~“(Y’) = &y&,,) ey”(y’)

=DhY(gL,h-‘)D~(h)e,“(y)~

= D&g> Kt”((y) F. (III. 12)

From (III. IO) and (III. 12) it is a simple exercise to derive an expression for
~Yy’~/ijx~. Thus,

ay’* ay” ay’”

-=---
axm axm ayD

= (g-‘%d K((y)g

= (g-‘a,g)“DgP(g-‘)Kr(y’)
= -(ga,g-‘)6 Kf(y’). (III. 13)
344 SALAM AND STRATHDEE

With the help of (III. 12) and (III. 13) one can easily extract the transformation rule
for the Yang-Mills gauge fields A,b(x) defined by (4.1), or

E,% Y> = 4%) WYY). (III. 14)

Since E,” is part of a contravariant vector it must transform according to

E,Y(x,~)-lE:Y(x,~‘)=E,‘(x,~)~
+ E/(x, y)$$. (III. 15)

Substitution of (111.14) into this rule gives

A$(x)qyy’) =Aa6(x)Kp”(Y)g+ E,“(x)ay’u 8X”

Since Ka“(y’) is a common factor in this equation, it may be removed. One is left
with the rule

A~h(x)=A,t(x)D.,b(g-‘)-E,“(x)(g~,g-1)8

or, equivalently,

~~(x)=g&(x)g-’ -g%J-‘9 (111.16)

where A,(x) = E,“(x)A,‘(x) Qb.

A more direct construction of the Killing vectors makes reference to infinitesimal
transformations. Thus, the infinitesimal form of the transformation rule L,, = gL,,h - ’
reads
L y+6y = (1 + dg”Q&) &,(I - Sh’Q,-),

i.e.,
dy’V,L, = 6g”QbL, - L,dh”Q,

to first order in the infinitesimals 6y’, 6g” and ah’. Multiply on the left by L,-’ and
use the definition of e,
iSyyle,,‘Qs = dgg(iLy-‘Q6LY - 6hzQ,
= Sg”Dbs(L,) Q6 - 6h”Q,-.

Extract the coefIicient of Q,

Sy”ern = &‘%a”&)
 ON KALUZA-KLEINTHEORY 345

since 6h” = 0. Hence the infinitesimal 6y’ is given by

W = &6D6”(L,) e,“(y)
= &sKs”(Y> (III. 17)

thereby defining the Killing vectors associated with left translations on G/H.
The same procedure could be followed for constructing the Killing vectors
associated with right translations. One should use the l-form e’= LdL -’ instead of e
to obtain Z?&” = DcB(L,-‘) ZB”. For the special case, H = 1, the result is particularly
simple. One finds Kg@= --ehtt.

APPENDIX IV: HARMONIC EXPANSIONS

The general form of harmonic expansions on the manifold G/H was discussed in
Section III. In particular, for the functions vi(x, y) which transform under the
combined left translations, y + y’, and associated tangent space rotations, h, such that

WiCx3Y> + V;txv V’) = IDijth) VjCx, V> (IV.1)

the appropriate expansion would be

(IV.2)

The sum includes all irreducible representations, D”, of G which contain D, on

restriction to the subgroup, H. If D, occurs more than once in D”, then the
supplementary label [ is needed. The matrices Dq; have dimension d, . The expansion
coefficients I/’ are fields on 4-dimensional spacetime which belong to the irreducible
representation D”,

They can be projected from w,(x, y) by integrating over the manifold

(IV.4)

Differentiation with respect to yW of the fields v,(x, y) is equivalent to an algebraic

operation on the components v”(x). The simplest covariant derivative that incor
porates ~3/3y“ would be

(IV.5)
346 SALAM AND STRATHDEE

In this formula we have used the expression

B u~4yl= euLei%qy9

which is one of the possible spin connections on G/H, derived in Appendix II.6 We
have also used the formula (3.17) which fixes the embedding of H in the tangent
space group, SO(N),
- ~c,~~~D(Z~~ = D(Q,).

Differentiation of the boost L,-’ leads back to the definitions (3.3) of e, viz.

apLy-’ = -L,-+,L,L,-’
= -e,6Q~L,-‘,

and this implies that, in any representation,

B,D”(L-‘) = -e,BD”(Q& D’(L,-‘). (IV.6)

On comparing this with (IV.5) one finds for the covariant derivative of D”,

V,D,;(L -‘) = -Di;(Q,) DJL -‘). (IV.7)

This formula was used in Section IV to reduce some terms from the 4 + N-
dimensional Lagrangian to 4-dimensional form.
Finally, the expansion formulae are sketched for two relativity simple cases.
(1) SU(2)/U(l). Th is is the case of spherical harmonics on the 2-dimensional
sphere. With G = SU(2) generated by Qnb = -Q,,, a, b = 1,2, 3 and H = U(1)
generated by Qi2 we can take y” = (0, p) with
L = ,-wQIz~-~QN (IV.8)
Y

The covariant basis vectors are contained in the l-form

= -dp cos t9Q,, + dcpsin t9Q,, - d8Q3,. (IV.9)

Expressed in matrix notation, with rows and columns indicated,

13 23 12
I91 0 0
e ‘=
u (IV. 10)
v, ( 0 sin 8 --cose 1 *
The last column gives the components of err’*(= e,,’ in the notation of Section III).
6 This spin connection corresponds to non-vanishing torsion on G/H. In general the spin connection
will contain terms additional to this one, including fluctuation terms, for example.
 ON KALUZA-KLEIN THEORY 341

The tangent space group, O(2), in this case coincides with H = U(1) so the
embedding problem is trivial. The 2-vector 9, resolves into helicity L = f 1 com-
binations.

and likewise for higher rank tensors. For a function, $n(x,v) of helicity II, integer or
half-integer, with respect to the tangent space group we have the expansion

(IV. 11)

where j takes the values IL (, I,?( + l,... and D.’ denotes the 2j + l-dimensional
representation of SU(2),

D,(i(L,,-‘) = D.<f(eBQ31evQ12)
= d,[j,(B) eimw, (IV.12)

in the notation of Wigner.

Generalization to the N-dimensional sphere SO(N + l)/SO(N) is straightforward
in principle though, of course, complicated in detail. There is no embedding to be
done since H = SO(N) coincides with the tangent space group. Our second example
illustrates a case where this is not so.
(2) SU(3)/SU(2) x U(l). With G = SU(3) generated by the octet of charges,
Qab = -Qb” (Q,’ = 0), a, b = 1,2,3 and H generated by the subset Q,“, a, /I = 1, 2
we can parametrize G/H by a complex 2-vector
L = ey’Q,‘+y2Q+h.c. (IV.13)
Y

This would not be the most convenient choice for practical computations but we do
not intend to present more than a sketch here. The question of interest is how to
embed H = SU(2) X U(1) in the tangent space group, SO(4). The general formula for
this, (3.17), could be applied but it is simpler in the present case to work ab initio.
The triplet of SU(3) decomposesunder the subgroup SU(2) x U(1) such that

3=2,,, + I-,,,, (IV. 14)

i.e., into an SU(2) doublet with U(1) quantum number l/3 and a singlet with -2/3.
For the octet the decomposition reads

8=1,+3,+2,+2-,. (IV.15)

If the octet is real then the doublets 2, and 2- 1 are related by complex conjugation.
In particular, the generators Qa3, Qju associated with G/H are in the representations
2, and 22,. respectively. According to the general principle expressed in formula
348 SALAMAND STRATHDEE

(3.17), the 4-vector of SO(4) - SU(2) x SU(2) must have the SU(2) x U( 1) content
of G/H,
(2,2)=2,-2-,. (IV.16)

This means that the embedding must be such that

(2, I)= 20,

(1,2)= 1, + l-1, (IV. 17)

for the 2-spinors of SO(4). An explicit construction for the SU(2) x U(1) decom-
position of a 4-vector, d,, is easily arranged. From the 2 x 2 matrix

h + 44 iM - 4h) (IV.18)
( W + 44) h - 44 1
select the first column. This belongs to the representation 2,) and so we define the
spinor #n with components

d-l,2=iWl-B2).
fi
(IV. 19)

Now the representation 2, is found in the SU(3) multiplets 8, 10, 27,..., with triality
zero. It follows, therefore, that the harmonic expansion of the 4-vector, 4,(x, y), must
take the form

h.(XPY) = n=,,E ,,... &ipwl)h”o. (IV.20)

where the coefficients 4p”(~) are complex. Expansions for tensors of SO(4) can be
constructed by straightforward generalization of this procedure. It is interesting to
note, however, that spinors of SO(4) cannot be represented in this way. This is clear
from the fact that the SU(2) x U(1) content of the 2-spinors, viz. 2,, 1, and 1 -, is
not to be found in any SU(3) multiplet. In fact, it is not possible to define spinors on
the manifold SU(3)/SU(2) X U( 1).

APPENDIX V: SPECTRUM OF 4 + 1 KALUZA-KLEIN THEORY

Since the ground state of the K = 1 theory is flat, it can be obtained as a vacuum
solution of the 4 + l-dimensional Einstein equations without a cosmological term.
For K > 1 the manifold G/H has constant curvature and it is necessary to introduce
non-geometrical (matter) field to act as sources. Because of the simplicity of the
 ON KALUZA-KLEIN THEORY 349

K = 1 theory one can easily obtain its excitation spectrum and relate it to an
underlying non-compact symmetry, SO( 1,2).
From the fact that the 4 f l-dimensional Einstein theory has five degrees of
freedom one expects that all massive states belong to multiplets of spin 2. This is
easily verified. Write the metric tensor in the form

g,, = VMN + k,m (V.1)

where M, N = 0. 1, 2, 3, 4 and r,~,,,,,,is the flat metric,

‘I.MN = diag(+l, -1, -1, -1, -1). (V.2)

Treating the components h,, as small quantities, one obtains the second order terms
in the Einstein Lagrangian

where J,, is an external source. The connections vanish so that simple partial
derivatives are indicated, h,,,N,L = a, hM,,,. The equations of motion derived from (V.3)
take the form

h MN,LI. -h ML,NL -h NL,Ml. + hLL,,N + n,AL,m. - rlndhi.LL = Jw. (V.4)

These equations are compatible only if J is conserved,

J MN,N -- 0. (V.5)

This is a relic of the Sdimensional general covariance. The usual way to solve (V.4),
subject to the compatibility condition (V.S), is to impose a co-ordinate condition. For
example, in the gauge
hMN,N -0
- (V.6)
they are solved by

h,,=$ [J,,--f (12,*-vj J,,I$

where 8’ represents the Sdimensional d’Alembertian.

On substituting the solution (V.7) back into the Lagrangian (V.3), discarding total
derivatives, one finds that Yz reduces to

(V.8)
350 SALAM AND STRATHDEE

This expression represents the effective interaction between conserved (i.e., physical)
sources due to the exchange of the Kaluza-Klein particles.
Now consider the pole, C? = 0, in (V.8). The residue simplifies on using the conser-
vation of JMN. Firstly, if a4 # 0 we can choose a frame in which

a,=a, and ai=o, i= 1,2,3. (V.9)

In this frame we have
Jw = Jw M=O, 1,2,3,4 (V.10)

SO that JMM = -Jii and JMi = Jij. It follows that, in the neighbourhood of the pole,
the effective interaction (V.8) reduces to

(V.11)

where the physically significant part of the source is the traceless 3-dimensional
tensor
Jim= Jij- fs,J,,. (V.12)

This result indicates that the massive states (a, # 0) are indeed purely spin 2. Their
masses are given by the eigenvalues of -ai = (n/2&)*, n = 1,2,..., where R is the
radius of the Kaluza-Klein circle.
On the other hand, if a4 = 0 we have the massless states and they can be analysed
conveniently by choosing a frame where

a,=a, and a, = a2=a4 = 0. (V.13)

In this frame the effective interaction reduces, in the neighbourhood of the pole, to the
form
(V.14)

where I refers to the O(2) helicity and

J, = Jc,, + iJ42,

J, = L (J, 1 + J,, - 2Ju)r

6
J-, = J4, -iJ42,

J-2=&J,,--2,)-G (V.15)
 ONKALUZA-KLEINTHEORY 351

The terms in (V.14) correspond to the exchange of graviton, photon and

Bands-Dicke scalar.
Turning now to the non-compact symmetry aspect of this theory, we note that the
general co-ordinate transformations in 4 + l-dimensions include, in particular, the set
6x” = 0,

(V.16)

where h4= 1/2zR. (The finite transformations are discussed in Appendix I. See, in
particular (1.25).) The infinitesimal generators are given by

Q, +%,. Q,=-;a,.. (V.17)

and these are easily seen to generate the algebra of SO( 1,2),

IQ,.Q,l=-Q,, IQ,, Q,l= Q,? IQ,, Q,l= Q2. (V.18)

Moreover, the Casimir invariant vanishes,

qabQaQb=- (?a,)‘- (vay)’ + (;ay)

= 0.

This means that fields defined on the circle belong to an irreducible representation of
SO( 1,2). It is infinite dimensional but not unitary, however. Since the mass operator
is given by --au2 = M’Q: one sees that the spectrum of masses is simply the spectrum
of Q: in the irreducible representation of SO(1,2).
Generalizations to higher dimensional manifolds are possible. For example, on the
sphere SK, which is invariant under SO(K + l), it is possible to define the action of
SO(1, K + 1). However, to find the excitation spectrum and perhaps relate it to the
spectrum of some combination of generators of SO( 1, K + l), it would be necessary
to include the excitations of the matter fields with are needed for the compac-
tilication.

ACKNOWLEDGMENT

The authors wish to thank S. Randjbar-Daemi and W. Mecklenburg for many stimulating and helpful
discussions.
352 SALAM AND STRATHDEE

REFERENCES

1. TH. KALUZA, Sitz. Preuss. Akad. Wiss. Berlin, Math. Phys. K1 (1921), 966.
2. 0. KLEIN, Z. Phys. 37 (1926), 895.
3. J. SCHERK AND J. H. SCHWARZ, Phys. Lett. B 57 (1975), 463; E. CREMMER AND J. SCHERK, Nucl.
Phys. B 103 (1976), 393; 108 (1976), 409.
4. B. DE WATT, “Dynamical Theory of Groups and Fields,” Gordon & Breach, New York, 1965.
5. Z. HORVATH, L. PALLA, E. CREMMER, AND J. SCHERK, Nucl. Phys. B 127 (1977), 57; L. PALLA,
“Proc. 1978 Tokyo Conf. on High Energy Physics,” p. 629.
6. Z. HORVATH AND L. PALLA, Nucl. Phys. B 142 (1978), 327.
7. P. G. 0. FREUND AND M. A. RUBIN, Phys. Len. B 97 (1980), 233.
8. A. S. SCHWAR~, Comm. Math. Phys. 56 (1977), 79; V. N. ROMANOV, A. S. SCHWARZ, AND Yu. S.
TYUPKIN, Nucl. Phys. B 130 (1977), 209; V. N. ROMANOV, I. V. FROLOV, AND A. S. SCHWARZ,
Teor. Mat. Fiz. 37 (1978), 305; A. S. SCHWARZ AND Yu. S. TYUPKIN, Nucl. Phys. B 187 (1981),
321; R. CASALBUONI, G. DOMOKOS, AND S. KOVESI-DOMOKOS, Phys. Rev. D 17 (1978), 2048.
9. J. F. LUCIANI, Nucl. Phys. B 135 (1978), 111; E. WIT~EN, Nucl. Phys. B 186 (1981), 412.
10. R. KERNER, Ann. Inst. H. PoincarP 9 (1968), 143; Y. M. CHO AND P. G. 0. FREUND, Phys. Rev. D
12 (1975), 1711; Y. M. CHO, J. Math. Phys. 16 (1975), 2029; Y. M. CHO AND P. S. JANG, Phys.
Rev. D 12 (1975), 2789; A. TRAUTMAN, Rep. Math. Phys. 16 (1975), 2029; K. I. MACRAE, Phys.
Rev. D 18 (1978), 3737; C. ORZALESI, Multidimensional unified theories, Parma preprint IFPR/TH
63 (1980); W. MECKLENBURG, Phys. Rev. D 21 (1980), 2149.
11. K. NOMIZU, Amer. J. Math. 76 (1954). 33; S. KOBAYASHI AND K. NOMIZU, “Foundations of
Differential Geometry,” Vol. II, Wiley-Interscience, New York, 1969.
12. See, for example, ABDUS SALAM AND J. STRATHDEE, Phys. Rev. 184 (1969), 1750; 184 (1969),
1760.
13. J. SCHERK AND J. H. SCHWARZ, Nucl. Phys. B 153 (1979), 61.