Kaluza-Klein compactications, supersymmetry, and

Calabi-Yau spaces: I
Andrew Strominger

I Introduction
As we have seen in d'Hoker's lectures, study of the ten-dimensional heterotic string dynam-
ics leads us to a low energy eective action of the form:
Z q  0 
1
Se [g; A; ; : : :] = 22 10 d x det(g )e
10 2 
R(g ) + 30 TrFA +



2
X
for a metric g of signature ( ; +; +;


; +) on a ten-dimensional manifold X 10, a connection
A on a principal G-bundle over X 10, a scalar eld , called the dilaton eld, and other elds
including the supersymmetric partners of these that we have not written here. Here R(g )
is the scalar curvature of g , G is compact, semisimple, FA is the curvature of A and TrFA2
represents a G-invariant metric on the Lie algebra of G. The study of the heterotic string
shows that it must be the case that the Cartan matrix of the Dynkin diagram for G is an
even unimodular lattice in dimension 16. The only two possibilities for G are E8  E8 or
Spin(32)=(Z=2Z) for a central Z=2Z which is not in the kernel of the projection to SO(32).
At the level we are working in these lectures, the delicate distinction between various
groups with Lie algebra SO(32) cannot be seen. For this reason, and also to simplify the
 Notes by John Morgan

1
notation, we shall consider SO(32) instead of the more exotic form. Lastly, the expression
p
d10x det(g ) is the volume element associated to the metric. The minus sign arises from
the fact that the metric is of signature (1; 9) so that the sign is necessary to have a positive
quantity under the square root.
Our goal in these lectures is to study and analyse this Lagranian in its own right, not
considering the small coreections arising from string theory. It turns out that, because of
the topological nature of the analysis, our conclusions remain largely valid in the exact
string theory. In fact, this Lagrangian was written down as a Lagrangian for supergravity
before string theory was invented, and so is of interest in its own right. There are two basic
problems that any such analysis should resolve if this theory is to be a reasonable model of
nature.
Problem 1. This action exists for metrics, connections, etc. on a ten-dimensional manifold
X 10, yet it seems fairly apparent that any theory of nature must have to do with four-
dimensional space-time of our universe. There must be an explanation of the `collapsing'
of ten space-time dimensions to four.
Problem 2. The two gauge groups that occur for this action, E8  E8 and SO(32), are not
appropriate ones for the observed matter content of the universe. Somehow by a mechanism
of spontaneous symmetry-breaking, we must get down to a smaller group which is suitable
for describing observed matter.
Our approach will be to rst look for classical ground states (solutions) minimizing the
action and study their features. At a later stage one should consider quantum corrections.
Motivated by the two problems we listed above, here are the properties that we want
our ground states (or classical solutions) to have:
1. The eective theory looks four-dimensional instead of ten-dimensional.
2. The gauge group G should be broken down to a `good' gauge group, one that ts with
the GUT (grand unication theory) scheme, so that it produces matter content observed in
2
nature. The gauge group of the standard model is SU (3)  SU (2)  U (1) and accounts for
all the matter that we know of, in the sense that all the `elementary' particles are grouped
together into a representation of this reductive group. The GUT schemes are to embed
this group in a larger group, SU (3)  SU (2)  U (1)  G, so that the representations that
occur in the standard model occur among those arising in the restriction of an irreducible
representation of G to the subgroup SU (3)  SU (2)  U (1). This scheme (among others)
allows the following set of groups for G:

SU (3)  SU (2)  U (1)  SU (5)  SO(10)  E6 :

Of course, on the level of groups on can continue the series with

E6  E7  E8 ;

but the irreducible representations that we must use have to be chiral and hence not iso-
morphic to their duals. This rules out E7 and E8 where the relevant representations are
real. Somehow we must break either E8  E8 or SO(32) down to one of these groups, most
likely E6.
3. Our solution should have N = 1 supersymmetry in dimension 4 (the N = 1 super
Poincare group). We do not observe in supersymmetry in nature, so whatever super-
symmetry exists in our eective four-dimensional gravity theory must be capable of being
entirely broken to N = 0, hopefully by small quantum corrections, which we shall not dis-
cuss here. There are no mechanisms for breaking N  2 supersymmetry down to a chiral
theory. This implies that we can have at most N = 1 supersymmetry in our model. On
the other hand, we desire nontrivial supersymmetry in our model because it is the best
mechanism we know which is capable of accounting for the very small ratios of masses of
the observed low energy particles, i.e., for solving the `hierarchy problem' as it is known to
physicists.

3
4. Our solution should yield a realistic matter content.

The story that we shall describe was worked out in the mid 1980's (see [Candelas,
Horowitz, Strominger, and Witten, Nucl. Phys. B 258 (1985)] and [Green, Schwarz,
Witten: Chapters 12 { 16]). More recently, with the advent of dualities, there are new
wrinkles, but the basic story as it was understood ten years ago remains relevant.

II Kaluza-Klein Model
Let us begin with a much simpler and much older example, which goes back to the work
of Kaluza-Klein in 1926. We consider the ve-dimensional Einstein-Hilbert action
Z q
1
S [gb] = 22 d5z det(gb)R(gb):
Y
(The variable z is real.) The action is for a metric gb on a ve-dimensional space, and
R(gb) is the scalar curvature of g . Lastly,  is Newton's constant. (The hats are simply to
distinguish this from other metrics which will arise later.) The classical equations of motion
for this action are Einstein's equations
RMN (gb) = 0;
the vanishing of the Ricci curvature tensor of gb.
One solution bg is on the manifold Y = R4  S 1 where the metric is
gbMN (z)dz M dz N =  dxdx + r2d2 ; where z = (x; ):
Here we have introduced several conventions which we shall use consistently. The indices
;  range from 0 to 3 and are indices for coordinates x in Minkowski four-dimensional
space time. The metric tensor  is the standard diagonal form ( ; +; +; +). The indices
M; N range from 0 to 4 and index the coordinates zM on Y . Here, r is a constant which
determines the radius of the internal circle.
4
 We begin with this basic solution and study the situation near this solution. It is
convenient to write metrics (or symmetric two-tensors) in a slightly dierent basis:
 
gbMN (z)dz M dz N = e =3 e (d + A dx)2 + g dxdx :

In this description the metric on the ve-dimensional space has become three elds: a four-
dimensional metric tensor g , a one form on four-space A dx , and a scalar eld . Of
course, at this point all these elds depend not only on the point in the four-space but also
on the auxiliary (or internal) variable .
Now let us consider the expansion in Fourier modes with respect to the variable  on
the circle:
1
X
g (z) = g ;n (x)ein ;
n= 1
X1
A (z) = A;n (x)ein ;
n= 1
X1
(z) = n(x)ein :
n= 1
At this point what we have done is to rewrite the ve-dimensional metric as an innite
collection of purely four-dimensional elds (metrics, one-forms, scalars). As we shall see
shortly, when taking the low energy eective limit we can ignore all but nitely many of
these elds. (We keep only the zero modes in the above Fourier expansions.)
Now we are ready to consider excitations (or linearized uctuations) around our solution
bg. Consider a general metric gbMN (z )dz M dz N near gb. The linearized uctuation is

h~MN (z) = gbMN (z) gbMN :

It is convenient to dene

hMN = h~MN 12 gbMN h~MN bg MN :

5
 We view the symmetric two-tensor h = hMN (z )dz M dz N as parametrizing the uctua-
tions of the metric. We must take some care here for there are gauge symmetries, the group
of dieomorphisms of Y . Metrics lying on the same orbit of this group action must be con-
sidered as equivalent since they dier simply reparametrizing the space. The standard way
around this problem is to pick a gauge (a slice perpendicular to the orbit of the gauge sym-
metries). In the current context the simplest gauge to pick is called the transverse gauge.
It consists of considering metrics satisfying

rM hMN = 0:
Now, as we observed above, the equations of motion are the Einstein equations

RMN (gb) = 0:
Expanding RMN (bg + h) = 0 to rst order in h yields
 
2b5g hMN (z) =  @ @ + r12 @2 hMN (x; ) = 0:
Combining this equation with the Fourier mode expansion we see that

2bg4hMN ;n(x) nr2 hMN ;n(x) = 0:

2

That is to say hMN ;n(x) satises the wave equation in four-space for a eld of mass jnj=r.
Invoking a basic result from quantum mechanics we know that the lowest energy excitation
of a eld of mass jnj=r has energy at least jnj=r. Now it is time to say more about the
size of the circle factor in Y . So far it has been a free parameter in the theory. We shall
take r to be our roughly the Planck length, which is 10 33 cm. At this length scale, all the
excitations of hMN ;n (x) have huge energy provided that n 6= 0. Since we are interested in
the low energy eective theory, we can safely ignore all the modes hMN ;n(x) except the zero
modes. Thus, at energy scales below 1=r, our eective theory is purely four-dimensional.
It has three elds on four-space which are now independent of the internal variable . The
6
elds are: a metric g0 = g ;0 (x)dxdx on R4 , a one-form A0 = A;0 (x)dx on four-space
and a scalar eld 0(x) on four-space.
Next, let us plug these elds into the Kaluza-Klein action and write the resulting eec-
tive four-dimensional action. It is actually a somewhat complicated, if routine, computation
to carry out the substitution; what comes out is:
Se [g0; A0; 0] =
Z   
p 1 1   1 1
= r 4 d x g0 2 R(g0) 4 e F F0 62 (r0) + (r1) + r2 1 +


;
4 0 0 2 2 2
R
where F0 dx dx is the curvature of the connection A . All the terms enclosed in the
0
last pair of parentheses represent massive particles and can be ignored in our low energy
eective action. Thus, we are left with
Z p 
Se [g0; A0; 0] = r 4 d4x g0 12 R(g0) 14 e0 F0 F0 61 2 (r0)2 :

R
We began with a ve-dimensional gravity theory and for suciently small radius r we
have found the low energy eective four-dimensional action and written this action in terms
of three four-dimensional elds: a metric, a connection which is related to the Killing vector
eld on the ve-manifold tangent to the circle directions, and a scalar eld  which is related
to the free parameter r in the ve-dimensional theory.
Let us now consider the symmetries of this eective four-dimensional action. We claim
that these are simply the symmetries of the original ve-manifold Y = R4  S 1 which
commute with the natural S 1-action coming from this product structure of Y . This group
is generated by two types of dieomorphisms. The rst are simply dieomorphisms of R4 .
These are generated by vector elds  on R4 , independent of . It is easy to see that the
action of these on our elds are given by:
 g = L (g );
 A = L (A);
7
   = L ():
Here, in all cases L denotes the Lie derivative in the  -direction. The second type of
dieomorphisms are those generated by vector elds of the form

x 7! x ;  7!  +  5 (x):

Under these vector elds we have

 (A ) = 1 @  5 ;

 g =   = 0:
Notice, in particular, that A = A dx transforms like an abelian (i.e., U (1)) gauge eld.
Of course, there are other dieomorphisms of the ve-manifold Y , which do not preserve
the S 1-action. But these dieomorphism mix the massive and low energy modes. Since we
have thrown away the massive modes in forming our eective four-dimensional Lagrangian,
these symmetries are broken in the limiting process, leaving the unbroken group for the
four-dimensional eective action exactly the one described in the previous paragraph.

III Compactifying Einstein's Equation from ten dimensions

to four dimensions
Now let us consider ten-dimensional pure gravity. The action is
Z q
S [gb] = d10x det(gb)R(gb);
X 10

where gb is a metric on X 10 with signature ( ; +;


; +). The classical solutions extremizing

this Einstein-Hilbert action are Ricci- at metrics

RMN (gb) = 0:

8
In analogy with the Kaluza-Klein model, let us study ground states (classical solutions) of
the following form: The manifold X 10 is a product R4  K 6 where K 6 is compact, and the
metric is of the form gb0 = (gb0)MN dz M dz N where
 0 !
(gb0)MN = ;
0 (g0)mn (y )
where g0 is a positive denite metric on K 6 . Our notation is similar to the ve-dimensional
case. The indices ;  run from 0 to 3 and index the coordinates x on Minkowski space.
The indices m; n index the internal coordinates y m which are local coordinates on K . The
indices M; N run from 0 to 9 and index local coordinates z = (x; y ) on the product R4  K 6 .
The split metric gb0 is Ricci- at if and only if g0 is a Ricci at metric on K 6. The existence
of such a metric, of course, imposes a topological condition on Y . We x now a classical
solution of this form.
As in the Kaluza-Klein model, let us study small uctuations about this classical solu-
tion:
hMN (z) = gbMN (z) (gb0)MN :
Again, there is the gauge symmetry of the dieomorphism of X 10 to deal with, and again
we use the transverse gauge. The Ricci- at condition to rst order in hMN is
 
24 + 26 hMN (z) = 0;
where
24h =  @ @ h
and
26(hmn) = (g0)abrarbhmn + 2Rambn hab:
Following the same line of reasoning as before, we expand in Fourier modes in harmonics
of K 6 . We write
X
h (z) = h ;k (x)Yk (y )
k

9
where
26Yk (y) = k Yk (y):
Here since the metric on Y is positive denite, this D'Alembertian on scalar functions on
Y and is the (negative of the) usual Laplacian. It is a non-positive operator, so that k  0
for all k. The equation satised by h ;k is

(24 k )h ;k (x) = 0:

Once again we nd that h ;k is a solution to the wave equation in four-space for a eld of
mass k . If we scale the metric on K 6 by multiplying it by 1=r, then k scales by 1=r2. In
particular, if we scale by 1=r for r very small, then for all k for which k 6= 0, all excitations
of the eld h ;k (x) become extremely massive, and hence can be ignored in our low energy
eective action.
There is a similar analysis for the elds hmn (z ). We impose the transverse gauge
condition on the metric in the K 6-direction and we expand the equations of motion in the
harmonics of K 6 :
X
hmn (z) = 'k (x)Ymn;k (y );
k
where this time Ymn;k (y )dy mdy n
are sections of the symmetric square of the cotangent
bundle of Y which are eigenvectors with eigenvalue 0k for the D'Alembertian:

26Ymn;k (y) = gabrarbYmn;k (y) + 2Ramb nYab;k (y) = 0k Ymn;k (y):

Thus, we obtain
24'k (x) 0k 'k (x) = 0:
As before, this is the wave equation for a wave of mass 0k .
Here, though, it is not necessarily the case that 0k  0 for all k. There can be a nite
number of negative eigenvalues for the operator on Y . A negative eigenvalue means that

10
our solution g0 is a critical point for the Einstein-Hilbert action for positive denite metrics
on Y which is not a local minimum for this action, i.e., an unstable critical point. We wish
to avoid this possibility, so for now we simply assume that our operator 26 on Y has no
negative eigenvalues. (This will be automatic by supersymmetry in the Calabi-Yau case.)
As long as 0k > 0, all excitations of this eld will have mass on the order of 1=r2 and
can be ignored in our low energy limit. Thus, we keep only those 'k (x) for which 0k = 0,
i.e., we keep only those 'k (x) for which the corresponding Ymn;k (y )dy mdy n is a harmonic
section of the symmetric square of the co-tangent bundle of K 6 . This space of harmonic
forms is the formal tangent space at g0 to the moduli space of Ricci- at metrics on K 6 .
So, for each parameter in the moduli space of solutions to the Einstein equations on the
internal space K 6 we have a massless scalar for the eective action on four-space. These
elds are the analogues of the massless eld  we found in the Kaluza-Klein model, a eld
which came from the one parameter, the radius, describing the metric on the internal space
in that theory.

IV Adding Matter to the Mix

We have made a fairly complete analysis of pure gravity compactied from ten dimensions
to four dimensions by using a compact six-dimensional manifold K 6 with a Ricci- at metric.
It is time now to add other terms to the action so as to create a theory that has matter
as well as gravity. Our goal here is to give the general nature of the terms that will add
matter, both bosons and fermions, to the action that we have been studying. In the next
section we will be more explicit as to the exact nature of the terms we are adding. Here
the analysis is quite general and will be sketched only since much of it is similar to what
we have done in great detail above for the metric.
First of all we can add boson elds by adding to the action a term of the form S (')

11
where ' is a scalar or vector-valued function on ten-dimensional space-time. Once again,
we expand ' in Fourier modes using the harmonics of Y , and we can disregard all but the
zero modes. In this way, after compactication ' becomes a boson eld on four-space.
Let us consider fermions. We x a spin structure on R4  Y 6 leading to spinor elds
which can be sections of the spin bundles tensored with other auxiliary bundles with
connections. The Cliord algebra for an inner product space of type (1; 9) is R[32], the
32  32 real matrix algebra. The even component of the Cliord algebra is the block 16  16
real matrices. This means that the spin bundle over Y has a real structure and decomposes
as SR+ (Y )  SR (Y ), each factor being a 16-dimensional real bundle. A section of SR+ (Y ) or
SR (Y ) is called a Majorana-Weyl spinor. Its chirality is by denition which of these two
spin bundles it is a section of. The metric gb0 leads to a spinor connection and thus to the
real Dirac operator
D=: (SR(Y )) ! (SR (Y )):
In terms of the gamma-matrices for Cliord multiplication we can write
D= = mDm
where Dm is the spin covariant derivative in the direction of the mth coordinate. The Dirac
equation on the ten-dimensional manifold is then written as
iD=(10) = 0:
Since the metric gb0 is split, so is the Dirac operator, and in fact the Dirac equation is
i(D=4 + D=6) = 0;
where D=4 and D=6 are the Dirac operators on Minkowski four-space and K 6 respectively.
The solutions to this equation can be written as a sum
X 0
(z ) = k (x) k (y )
k

12
with
iD=4 k (x) = 0k k (x)
and
iD=6 k0 (y ) = 0k k0 (y ):
Again, throwing away massive uctuations, means keeping only the zero modes of the Dirac
equation on the internal space K 6.

V The Eective action from ten-dimensional heterotic string

theory
The eective action that heterotic string theory in ten dimensions gives us is

Se [g; '; ; ; ; A; H ] =
Z q 
1
= 2 d x det(g )e
10 2' R(g ) + 4(r')2 13 H 2
2 Y 
0
+ 30 Tr FA M
2 MNP DN P D= Tr D= + (


) :
The elds in question are as follows: There is a metric g of signature ( ; +;


+) on Y .
There is the dilaton eld ' which is a real scalar eld. There is a connection A on a
principal E8  E8 or SO(32) bundle P over Y . Its curvature is denoted FA . The axion
form H is a three-form with the property that

dH = 0 (tr R ^ R 30 1 Tr F ^ F ):
A A

This means that we can write H at least locally in the form

 
H = dB + 0 1
CS (rg) 30 CS (A) ;
where B is a two-form and CS (A) is the Chern-Simons three-form of the connection A. The
boson elds g; B; ' are grouped together in a multiplet (i.e., they are various components
13
of an irreducible representation). This multiplet has a supersymmetric partner which is a
multiplet given by a pair of spinors ( ; ). The spinor is a Majorana-Weyl spinor with
values in the cotangent bundle of Y ; that is to say, it is a section of SR+ (Y )
T (Y ). It is
called the gravitino. The spinor  is a Majorana-Weyl spinor of opposite chirality to that
of . That is to say, it is a section of SR (Y ). It is called the dilatino. The gauge eld A
has its supersymmetric partner , called the gaugino. This eld is a Majorana-Weyl spinor
with values in the adjoint bundle of P and is of the same chirality as . That is to say, it
is a section of SR+ (Y )
adP .
Having discussed the elds that occur, let us turn to the functions of them that appear
in the action. As usual, R(g ) is the scalar curvature of g and TrFA2 is the norm of FA in
the G-invariant metric in the adjoint bundle. Lastly, MNP is obtained from the gamma-
matrices by skew-symmetrizing:

MNP = [M N P ]:

These are the gamma-matrices that give the representation of three-forms on spinors. In
language more familiar to mathematicians, all the spinor pairings in the action arise from
the two basic pairings: the nondegenerate inner product between spinors of the opposite
chirality:
h i: SR+(Y )
SR(Y ) ! R
and the adjoint of Cliord multiplication

SR+ (Y )
SR+ (Y ) ! T (Y ): (1)

(Here we are viewing Cliord multiplication as a pairing T (Y )

SR+ (Y ) ! SR (Y ). To write
its adjoint as we have above, we use the isomorphisms T (Y ) = T (Y ), given by the metric,
and SR+ (Y ) 
= SR (Y ) , given by the inner product.) The term D= in the action is simply
the inner product h; D=i. Likewise, TrD= is Trh; D=i dened from the inner product

14
on spinors and the trace in the adjoint representation. Lastly, the pairing M MNP DN P
in the action means the pairing h ; r i, where the pairing is between a one-form with
values in the positive spin bundle and a two-form with values in the same spin bundle. The
pairing is given by wedging together the forms to produce a three-form and then letting
that three-form operate by Cliord multiplication on one of the spinors to produce a spinor
of minus chirality which can then be paired with the plus chirality spinor to produce a real
number.
In addition, there are other terms (indicated by (


)) that we have not explicitly written
down and which are not important for the analysis that we shall make. These extra terms
come in two types. Some are required by supersymmetry. There are, for example, terms
which are fourth power in the spinor elds such as ( )2. There are also terms which are
higher order in 0 , which is a dimensionful parameter. These can be ignored if we assume
that, even though K 6 is on the Planck scale, it is much larger that the string length.
This action clearly has a group of symmetries: the group of orientation-preserving
dieomorphisms of the underlying manifold. But in addition, we have gone to great lengths
to build it so that it has supersymmetry, N = 1 local supersymmetry to be more precise.
For example, our boson elds form two multiplets (g; B; ') and A, and each multiplet has
its fermionic partner ( ; ) and , respectively. This of course, is only a necessary condition
for the existence of supersymmetry. But in fact the N = 1 local supersymmetry algebra
is generated by sections  of the Majorana-Weyl spin bundle SR+ (Y ). These generators are
odd elements in the super Lie algebra. The innitessimal transformation laws are partially
given by:

  = FMN MN  + (fermions2
 M = rM  41 HMAB AB  + (fermions)2
1H
  = ( M rM ') + 24 MNP  + (fermions)2
MNP

15
 gMN =  N M +  M N + (fermions)2
 AM =  M  + (fermions)2
 B =



  =

The complete laws can be written down, but we shall not do that here as we will not need
them.
The space of classical solutions is invariant under the supersymmetry algebra generated
by these innitessimal (odd) symmetries, and on the space of solutions the algebra generated
by these innitessimal symmetries closes to yield the N = 1 local supersymmetry algebra.
In fact, one has the following formula for the supercommutator:

[ ; 0 ] = v + T

where v is the vector eld obtained by pairing  and 0 under the pairing given in Equation 1.
Written in terms of Cliord matrices we have

vm = 0 m:

The term T is an expression which includes gauge transformations and terms which vanish
when one imposes the equations of motion.
All of this was done without worrying about the compactication. The next step in
the study is to show under what conditions we can compactify down to four-dimensions.
In doing this compactication much of the N = 1 local supersymmetry will be broken,
but in special circumstances we will be able to retain N = 1 super Poincare symmetry in
dimension four. This is where the next lecture will begin.

16