Supersymmetric Yang-Mills Theory On A Four-Manifold

Download as pdf or txt
Download as pdf or txt
You are on page 1of 67

IASSNS-HEP-94/5

February, 1994
arXiv:hep-th/9403195v1 31 Mar 1994

SUPERSYMMETRIC YANG-MILLS THEORY


ON A FOUR-MANIFOLD


Edward Witten

School of Natural Sciences


Institute for Advanced Study
Olden Lane
Princeton, NJ 08540

ABSTRACT

By exploiting standard facts about N = 1 and N = 2 supersymmetric Yang-


Mills theory, the Donaldson invariants of four-manifolds that admit a Kahler met-
ric can be computed. The results are in agreement with available mathematical
computations, and provide a powerful check on the standard claims about super-
symmetric Yang-Mills theory.

Research supported in part by NSF Grant PHY92-45317.


1. Introduction

In four-dimensional supersymmetric Yang-Mills theory formulated on flat R4 ,


certain correlation functions are independent of spatial separation and are hence
effectively computable by going to short distances. This is the basis for one of
the most fruitful techniques for studying dynamics of those theories [1,2], and
gains even more power when combined with analysis of instanton corrections to
superpotentials [3].

The facts used to establish such results on flat R4 also have global manifes-
tations, especially in the case of N = 2 and N = 4 super Yang-Mills theory.
For example, when N = 2 super Yang-Mills theory is formulated (in a suitable
twisted fashion which we will recall) on an arbitrary four-manifold M, one can
define certain correlation functions which are independent of the metric of M [4].
These correlation functions are thus, crudely speaking, topological invariants of
M. Actually, since the Yang-Mills action contains derivatives, its formulation re-
quires a differentiable structure on M. Hence, the special correlation functions are
invariants of M as a differentiable manifold. In fact [4,5], they are equivalent to
the celebrated Donaldson invariants [6,7], which are the basis of much of what is
known about smooth four-manifolds.

The realization of the Donaldson invariants as correlation functions of a physi-


cal theory leads one to wonder whether physical methods could be used to calculate
them, or conversely, whether topological computations can be used to constrain
the behavior of physical models. In this paper, I will present some results in this
direction. To be specific, I will show how standard properties of N = 1 super-
symmetric Yang-Mills theory can be used to determine the Donaldson invariants
of those four manifolds that admit a Kahler metric. This is an area that has been
much investigated (for reviews see [8,9]), and the formulas that we will obtain
agree with the known mathematical formulas. In fact, comparison with some of
those formulas especially the formulas of OGrady [10] for K3 surfaces and of
Kronheimer and Mrowka [11] for so-called four-manifolds of simple type helped

2
considerably in working out the argument presented in this paper.

As in a recent study of the Verlinde algebra and the cohomology of the Grass-
mannian [12], the basic idea will be to exploit the way a topological theory arises
as part of a physical theory. The topological correlation functions of interest will
arise as specific correlation functions in a model (essentially N = 1 supersymmetric
Yang-Mills theory) which has asymptotic freedom, chiral symmetry breaking, and
a dynamically generated mass gap; the mass gap in particular severely constrains
the topological correlation functions.

To make the subject as accessible as possible, the paper will be organized as


follows. The problem and its solution are described in a relatively non-technical
way in 2. Many technical details are deferred for 3.

There have been a couple of other recent papers considering supersymmetric


Yang-Mills theory on Kahler manifolds. Park [13] described many of the relevant
general features and also adapted the two-dimensional construction of [14] to four
dimensions where unfortunately this approach does not immediately give much
computational power. Johansen [15] formally wrote down the twisting of N = 1
models on Kahler manifolds.

2. Outline Of The Argument

First let us recall some of the basic properties of N = 1 and N = 2 super


Yang-Mills theory in four dimensions, with gauge group G. In the minimal N = 1
theory, without extra matter multiplets, the fields are a gauge field Am and a
gluino field (a Majorana spinor with values in the adjoint representation). The
Lagrangian is

Z  
1 4 1 mn m
L= 2 d x g Fmn F iDm . (2.1)
e 4

Our conventions are as in Wess and Bagger [16] and are explained at the beginning

3
of 3. Classically there is a U(1) R symmetry

ei
(2.2)
ei .

Instantons reduce the symmetry group to a subgroup Z2h of U(1), where h is the
dual Coxeter number of G; for G = SU(2), the main case that we will consider for
illustration, the symmetry group is Z4 .

This theory is asymptotically free, and its key properties are believed to be
those of conventional QCD: confinement, a mass gap, and spontaneous chiral sym-
metry breaking. The chiral symmetry in question is the finite group Z2h just
described. The standard conjecture is that this symmetry group is broken down
to a Z2 subgroup, whose unique non-trivial element is the transformation

, . (2.3)

This symmetry breaking produces an h-fold degeneracy of the vacuum, which is


believed to be the only degeneracy.

In fact, the symmetry in (2.3) is equivalent to a 2 rotation in space-time, and


so cannot be spontaneously broken if rotation invariance is unbroken. On the other
hand, additional unbroken symmetries would forbid fermion masses. The standard
conjectures stated in the last paragraph are thus in keeping with the experience of
ordinary QCD: in strongly coupled theories of gauge bosons and fermions only, the
unbroken symmetry group is a maximal subgroup of the global symmetry group
that allows fermion masses.

To this minimal N = 1 theory, it is possible to add matter fields in the form


of chiral superfields. Such a superfield is a bose-fermi pair = (, ), of spin
(0, 1/2), and transforming in an arbitrary representation of the gauge group. In
addition to the gauge interactions, the chiral superfields can have additional inter-
actions governed by a superpotential. The only special case of the superpotential

4
that will be important in this paper is the following: chiral superfields in a real
representation of G can have a mass term
Z
d4 x d2 m2 + h.c. (2.4)

compatible with supersymmetry.

An important real representation is the adjoint representation. If one adds


to the N = 1 theory a chiral superfield in the adjoint representation, then
the minimally coupled gauge theory (with vanishing superpotential) in fact has
N = 2 supersymmetry. The Lagrangian is written at the beginning of 3. It is
possible to add a bare mass term as in (2.4) for the multiplet, breaking N = 2
supersymmetry down to N = 1 supersymmetry.

Though the N = 2 theory is still asymptotically free, its behavior is quite


different from that of the minimal N = 1 theory. The chiral symmetry group is
now, at the classical level, a group U(2) acting on the pair (, ). Instantons break
this to a subgroup whose connected component is isomorphic to SU(2); we will
call this internal SU(2) symmetry group SU(2)I .

In contrast to what one might guess from the behavior of strongly coupled
gauge theories without scalars, it is believed that SU(2)I is not spontaneously
broken and that there is no mass gap. In fact, this follows from the form of the
classical potential
 2
V = Tr , , (2.5)

with a complex scalar in the adjoint representation. This potential has flat
6 0, breaking the gauge symmetry down to U(1)r (r being
directions, with hi =
the rank of G), and leaving a weakly coupled abelian theory, in which all charged
fields are massive. This behavior holds in the quantum theory at least in the weakly
coupled region of large hi, since N = 2 supersymmetry permits no perturbations
that would lift the vacuum degeneracy in the flat directions. The instanton contri-
butions in the weakly coupled theory for large hi have been analyzed by Seiberg
[17].

5
The goal in what follows is to use facts about the topology of four dimensions
to retrieve the conclusion just stated that the minimal N = 2 theory does not
have a mass gap, and conversely to use the mass gap and related properties of the
N = 1 theory to deduce conclusions about the topology of four dimensions.

2.1. Twisting

First we will recall how one constructs from the N = 2 super Yang-Mills theory
a twisted topological field theory. For more details, the reader can consult [4] and
also 3 below.

The rotation group K in four dimensional Euclidean space is locally SU(2)L


SU(2)R . In addition, the connected component of the global symmetry group of
the N = 2 theory is, as explained above, SU(2)I . The theory, when formulated on
a flat R4 , therefore has a global symmetry group

H = SU(2)L SU(2)R SU(2)I . (2.6)

The supercharges Q i and Qi transform under H as (1/2, 0, 1/2) and (0, 1/2, 1/2),
respectively.

Apart from its standard embedding in H (as SU(2)L SU(2)R ), K also


has nonstandard embeddings. Let SU(2)R be a diagonal subgroup of SU(2)R
SU(2)I , and let

K = SU(2)L SU(2)R . (2.7)

Then K is isomorphic to K, and we can think of the N = 2 super Yang-Mills theory


as a Poincare invariant theory with rotations acting by K . Use of K instead of
K to generate rotations is natural if one replaces the standard stress tensor T of
the theory by a modified stress tensor T (which differs from T by addition of a
derivative term that does not contribute to the translation generators).

6
But this substitution changes the physical interpretation of the theory consid-
erably. All fields, commuting or anticommuting, have integer spin with respect to
K . In particular, this is so for the supercharges, which under SU(2)L SU(2)R
transform as (1/2, 1/2) (0, 1) (0, 0).

The (0, 0) component of the supercharge is the rotation invariant object Q =


i Qi . By virtue of the supersymmetry algebra, Q obeys Q2 = 0; this follows
from the fact that Q2 transforms as (0, 0) while the only bosonic operators in the
supersymmetry algebra are the momenta, transforming as (1/2, 1/2). Moreover,
it is possible to find a component S of the underlying supercurrent such that the
modified stress tensor T can be written

T = {Q, S }. (2.8)

These facts ensure that with Q regarded as a BRST operator, and T used as
the stress tensor when coupling to gravity, the theory can be interpreted as a
topological field theory.

Indeed [4], more or less because Q is Lorentz invariant in the K sense, the
N = 2 theory can be formulated on an arbitrary four manifold M in such a way
as to preserve the existence of the fermionic symmetry Q and the basic relations
Q2 = 0 and (2.8). At this stage the twisted theory comes into its own. On flat
R4 , the twisted theory is merely a different way of looking at the usual physical
theory; but as soon as we formulate the theory on a curved space-time, the twisted
theory is really different, because of the use of a different stress tensor.

2.2. Operators

The important part of the BRST symmetry (in the sense that other fields do
not enter in the construction of observables) is

Am = im
m = Dm (2.9)
= 0.

7
Here A is the gauge field, m is a linear combination of the fermions (the details
will be described in 3), is a scalar, and is an anticommuting parameter. One
easily verifies from this formula that, up to a gauge transformation, Q2 = 0 in
acting on these fields. (The multiplet (2.9) has a mathematical interpretation in
terms of the equivariant cohomology of the gauge group acting on the space of
connections, but that need not concern us here.)

Since is BRST invariant, but cannot be written as {Q, . . .}, the obvious
BRST invariant observables are of the form P (), with P an invariant polynomial
on the Lie algebra of G. For G = SU(2), the basic invariant polynomial is the
quadratic Casimir operator. (All others are polynomials in this and lead to nothing
essentially new.) So the basic Q-invariant observable is
1
O(x) = O(0) (x) = 2
Tr 2 (x), (2.10)
8
with Tr the trace in the fundamental two dimensional representation of SU(2).
More generally, for a Lie group of rank r, there are r independent Casimir operators
and accordingly r independent operators generalizing O(0) (x).

So the first example of a topological correlation function is

hO(x1 )O(x2 ) . . . O(xs )i , (2.11)

with arbitrary distinct points xi . (There is no singularity in the O(x)O(y) operator


product for x y, so one could in fact let some of the xi coincide, but it is more
convenient to keep them distinct.) This correlation function is independent of the
metric of M because of (2.8); and so it must be independent of the choice of the
xi , because up to a change in metric on M, there is no invariant information in
the choice of distinct points x1 . . . xs M. To prove more explicitly that (2.11)
is independent of the xi , note that although O cannot be written as {Q, . . .}, its
derivative can be:
Tr m
m
O = {Q, }. (2.12)
x 4 2

I write this equation as dO(0) = {Q, O(1) }, where O(1) is an operator-valued

8
one-form O(1) = Tr m dxm /4 2 . It is the beginning of a descent procedure. One
recursively finds operator-valued k-forms O(k) such that

dO(k) = {Q, O(k+1) }. (2.13)

For full details on this, see [4]. (2.13) means that O(k) is Q-invariant up to an exact
form and is closed up to {Q, . . .}. Accordingly, if is a k-dimensional submanifold
of M (or more generally a k-dimensional homology cycle), then
Z
I() = O(k) (2.14)

is BRST invariant and depends only on the homology class of .

In practice Donaldson theory is most intensively studied on simply connected


four manifolds M. In that case, k-dimensional homology cycles exist only for
k = 0, 2, 4. For k = 0, is a point x M, and I() is simply our friend O(x).
For k = 4, must be M itself. For O constructed from the quadratic Casimir
operator, I(M) turns out to be simply an elementary topological invariant, the
R
instanton number I(M) = M Tr F F/8 2 . Therefore, for G = SU(2), interesting
quantum operators do not arise for k = 4. Interesting operators with k = 4 do
arise for other groups, which have higher Casimir operators.

Hence for SU(2), apart from k = 0, we focus on the case that k = 2 and that
M is an oriented two dimensional submanifold of M. For each such , we
get a BRST invariant operator I(), which up to {Q, . . .} depends only on the
homology class of . If we denote the components of the operator-valued two-form
O(2) as
1
Zmn (x) = Tr (Fmn im n ) (2.15)
4 2
then
Z
I() = Zmn d mn . (2.16)

9
The correlation functions

hO(x1 ) . . . O(xr )I(1 ) . . . I(s )i (2.17)

are the celebrated Donaldson invariants of smooth four-manifolds.

There is actually a small subtlety here: the Donaldson invariants as usually


defined by topologists are larger by a universal constant factor than the correlation
functions just introduced. This factor is the number of elements in the center of
the gauge group; we will call the center Z(G) and let #Z(G) denote the number
of its elements. Thus if h iT denotes the Donaldson invariants as usually defined
topologically, then

hO(x1 ) . . . O(xr )I(1 ) . . . I(s )iT = #Z(G) hO(x1 ) . . . O(xr )I(1 ) . . . I(s )i .
(2.18)
The reason for this is that to recover the conventional topological definition of
Donaldsons invariants from the path integral (by a semi-classical evaluation of the
path integral, discussed in [4,5]) one must delete a factor of 1/#Z(G) that appears
in the properly normalized path integral and is explained in [18, p. 159]. In fact,
in the Faddeev-Popov definition of the path integral, one divides by the volume
of the gauge group, taking account of the center of the group even though it acts
trivially on the space of connections; but topologists usually omit to divide by the
order of the center.

The Dimension Of Moduli Space

I will not attempt here a systematic account of the field theoretic approach to
Donaldson theory. However, one point is so fundamental that it requires mention.

The classical N = 2 theory has a U(2) symmetry, acting on the two fermi fields
( , ). So far we have exploited the anomaly-free SU(2), in twisting the model.
The center U(1) U(2) is anomalous; nevertheless it plays an essential role. Let
us call this quantum number U.

10
As explained in [4], on a given four-manifold and for a given instanton number,
the total violation of U, which we will call U, equals the dimension of the Yang-
Mills instanton moduli space M. For SU(2) this is

3
U = dim M = 8k ( + ) , (2.19)
2

with k the instanton number and and the Euler characteristic and signature
of M. (The quantity ( + )/2 is always integral.)

On the other hand, the operators O and I() have U = 4 and U = 2, respec-
tively, so a correlation function

hO(x1 ) . . . O(xr )I(1 ) . . . I(s )i (2.20)

vanishes unless
3
4r + 2s = dim M = 8k ( + ) . (2.21)
2

2.3. Role Of A Mass Gap

N = 2 supersymmetric Yang-Mills theory does not have a mass gap for


reasons explained earlier. Nevertheless, we will ultimately perform calculations by
perturbing to a situation in which there is a mass gap. To see how the mass gap
enters, we will first assume counter-factually that the N = 2 theory does have
a mass gap, and see what computations become possible.

Actually, some of the arguments (those that involve I() as well as O) require a
somewhat stronger assumption. We will assume that the ground state structure, on
any space-time manifold, is the same as it is in the bulk theory. This means roughly
that the vacua can be completely labeled by gauge-invariant local observables. The
only known theories with mass gap in which this is not so are theories with unbroken
and unconfined gauge symmetries, in which measurements of Wilson lines or global
holonomies are needed to distinguish the vacua. Thus the assumption that we will

11
exploit is essentially that the N = 2 theory has a mass gap, and no unbroken,
unconfined gauge symmetries.

We will also temporarily make a further simplification and assume that there
is only one vacuum. This is a minor point and will eventually be corrected by
introducing a sum over vacua.

Consequences Of The Assumption

With our assumptions, let us determine the Donaldson invariants. First we


consider the zero point function that is the partition function in the absence of
any operator insertions. We will use the symbol hXi to refer to an unnormalized
path integral on the four-manifold M with an insertion of the operator X, so the
partition function will be denoted as h1i. When we need to refer to the normalized
vacuum expectation value of X in the infinite volume limit (on a flat R4 ), we will
write this as hXi , with being the infinite volume vacuum state.

We use our liberty to pick a convenient metric on M. We begin with any


metric g, and then scale it up by g tg, with t a positive constant, and consider
what happens for t . The metric on M becomes everywhere nearly flat. In a
theory with a mass gap, the response to a background gravitational field is given
by an effective action that can be expanded as a sum of local operators. In fact,
the partition function is

h1i = exp(Lef f ), (2.22)

where Lef f has an expansion

Z

Lef f = d4 x g u + vR + wR2 + . . . , (2.23)

with constants u, v, w, . . . and local operators 1, R, R2 , . . . (R is the Ricci scalar of


M). This is the general structure that would arise in any local theory with a mass
gap, but in the case at hand the expansion is severely restricted by topological

12
invariance. The above expansion is valid for large t. If U is an operator of di-
R
mension n, then d4 x gU scales as t4n , so the terms written explicitly in (2.23)
make contributions of order t4 , t2 , and 1. Topological invariance means that in the
particular case at hand, the expansion must be independent of t, so only operators
of dimension four can appear. The only topological invariants of a four-manifold
that can be written as the integral of a local operator are the Euler characteristic
and the signature , so

h1i = exp(a + b) (2.24)

with some universal constants a and b. In the case of SU(2) Donaldson theory, we
will eventually determine a and b by comparing to computations on some particular
four-manifolds.

Now we include the local operator O and consider its correlation functions

hO(x1 ) . . . O(xr )i . (2.25)

Upon scaling up the metric, we can assume that the xi are far apart from one
another and inserted in a region of M that is essentially flat on a scale much larger
than any Compton wavelength in the theory. So via cluster decomposition, O can
be replaced by its vacuum expectation value hOi :

hO(x1 ) . . . O(xr )i = hOi r h1i. (2.26)

This can conveniently be written

hexp(O)i = h1i exp(hOi ), (2.27)

with a complex parameter.

13
The arguments become more interesting if we consider the operator I(). I
will first show that given the assumptions, the one point function of this operator
vanishes,

hI()i = 0, (2.28)

for every four manifold M and every two dimensional surface (or homology cycle)
M. In fact,
Z
hI()i = d mn hZmni, (2.29)

and the desired conclusion will follow from suitable properties of the one point
function

hZmn (x)i. (2.30)

Zmn (x) is a well-defined, gauge-invariant local operator of the sort one usually
studies in physical applications of Yang-Mills theory. Its one point function is
a standard sort of Yang-Mills observable but is not a topological invariant; only
integrated expressions like (2.29) are such invariants. In studying (2.30) as a pre-
lude to analyzing (2.29), we are taking the main step in this paper: constraining
the topological correlation functions by exploiting the way they arise as special
correlators in a physical theory.

Clearly, on R4 with a flat metric, hZmni = 0 by Lorentz invariance. To de-


termine what happens on a general four-manifold, we again consider an arbitrary
metric g scaled up by g tg with t . Since the area of grows like t2 for
t , to show that the integrated expectation value (2.29) vanishes, it suffices
to show that for every x M, hZmn (x)i vanishes for large t faster than 1/t2 . This
is so as the mass gap lets us write an expansion for hZmn (x)i in terms of local
invariants of the Riemannian geometry of M. The expansion looks like

hZmn i = Dm RDn Ds D s R . . . (2.31)

Since all possible terms have dimension considerably greater than two, the expec-

14
tation value vanishes faster than 1/t2 .

Now, we move on to compute the two point function of I(). Let 1 , 2 be


two two-dimensional submanifolds of M. The desired two point function is

Z
hI(1 )I(2 )i = hZmn (x)Zpq (y)i d mn (x)d pq (y). (2.32)
1 2

As before, we analyze this by scaling up the metric by g tg, with t . For


x 6= y, hZmn (x)Zpq (y)i vanishes faster than 1/t4 , by the same reasoning as above.
The only possible surviving contribution would be a contribution that for t
is localized where x = y, that is at the intersection points of 1 and 2 .

By perturbing 1 and 2 slightly within their homology (and even homo-


topy) classes, we can assume that they meet transversely at finitely many points
w1 , . . . , wn . We can also assume that the metric of M is flat in a neighborhood
of the wi and that the a look locally like intersections of coordinate hyperplanes.
The contribution to (2.32) which for t is localized at one of the intersection
points wi can only depend on local invariants of the behavior of the a near wi .
The only such invariant is the relative orientation with which 1 and 2 meet
at wi . On symmetry grounds, the local contribution in (2.32) is proportional to
this relative orientation. (The operator I() is defined by integrating a differential
form over and so changes sign if the orientation of is reversed; reversing the
orientation of one of the a changes the relative orientation with which 1 and 2
meet at wi .) So given our assumption the sum of the local contributions in
(2.32) is proportional to the algebraic intersection number of the s:

hI(1 )I(2 )i = #(1 2 ) h1i. (2.33)

Here is an unknown universal constant. This formula is valid even for 2 = 1 ;


one sees this by perturbing 2 within its homotopy class to a new surface 1 with
only transverse intersections with 1 .

15
By a generalization of this reasoning, if 1 . . . s are independent two-
dimensional submanifolds of M, then the mass gap implies

* !+
X X
exp a I(a ) = exp a b #(a b ) h1i. (2.34)
a
2
a,b

Indeed, the a can be assumed to have only pairwise intersections; the right hand
side of (2.34) comes from the local contributions of the intersection points. The
I(a ) are analogous to quantum fields with Gaussian correlations controlled by
the two point function (2.33).

No new phenomena occur if we consider correlators with O and I() together,


since the points at which O is inserted can be taken disjoint from the surfaces a
and then cluster decomposition can be used, as before, to replace the Os by their
vacuum expectation values. So we can combine and (2.27) and (2.34) to get

* !+
X X
exp a I(a ) + O = exp a b #(a b ) + hOi h1i.
a
2
a,b
(2.35)

Several Vacua

It remains to generalize this formula to take account of the possibility that the
theory has not one vacuum state but several. Assume existence of a discrete set
of vacua , with ranging over some finite set S. To generalize (2.35), we simply
write a sum over contributions of the , allowing for the fact that , hOi, and the
constants a and b of equation (2.24) may depend on the choice of vacuum. We get

* !+
X X X
exp a I(a ) + O = C exp a b #(a b ) + hOi .
a
2
S a,b
(2.36)
Here I have written the partition function h1i in the vacuum as C ; by (2.24) it

16
is of the form

C = exp (a + b ) (2.37)

with some universal constant a and b .

2.4. Restriction To Kahler Manifolds And N = 1

Upon comparison to mathematical computations of Donaldson invariants in


special cases, one finds that these formulas are in fact false. Therefore, the N = 2
theory does not obey our assumptions; this comes as no surprise, in view of the flat
directions in the classical potential which ensure that the N = 2 theory does not
have a mass gap. Were (2.36) valid, Donaldson theory would detect only familiar,
classical invariants of four-manifolds and so would not have its usual mathematical
applications.

However, in certain respects, the formulas are nearly true at least for G =
SU(2) where mathematical computations have been performed. For instance, they
are true (with minor modification) for K3 surfaces according to results of [10].
For a large class of four manifolds (those of simple type), (2.36) gives the right
asymptotic behavior for large values of a , according to results of Kronheimer
and Mrowka [11]. (To be more precise, according to [11], the Gaussian function
of the a that appears on the right hand side must be multiplied by a sum of
exponentials, which are subleading but contain all of the subtle information in
the theory.) Why does our simple computation based on a faulty assumption enjoy
these successes?

The thesis of the present paper will be that this is so because the assumptions
are valid for N = 1 super Yang-Mills theory and this is almost as good. To be
more precise, in what follows we will show how N = 1 super Yang-Mills theory
can be used to study Donaldson theory in the case that the four-manifold M is a
Kahler manifold (that is, a manifold admitting a Kahler metric). This is in fact a
very basic case in Donaldson theory and the main arena of actual computations.

17
For some time it was conjectured that the Donaldson invariants are zero except for
four-manifolds admitting a Kahler metric; this is now known to be false [19], but
the known counter-examples are still closely related to Kahler manifolds.

The Twisted Theory On A Kahler Manifold

For our purposes, the definition of a four-dimensional Kahler manifold is that


it is a Riemannian four-manifold on which the holonomy group is not the usual
SU(2)L SU(2)R , but is contained in SU(2)L U(1)R , with U(1)R being a sub-
group of SU(2)R isomorphic to U(1).

The two dimensional representation of SU(2) has a single anti-symmetric in-


variant; given two doublets v , w this invariant can be written as v w . Upon
restriction to a U(1) subgroup, the same representation also admits a symmetric
invariant, which we might write as v w . The implication of this for the twisted
N = 2 model is as follows. While on a general four-manifold the twisted theory has
a single BRST operator Q = i Qi , on a Kahler manifold there is also a second
one Q = i Qi .

One way to formulate this is that on a Kahler manifold, one can write

Q = Q1 + Q2 , (2.38)

where Q1 is the part of Q inside one N = 1 subalgebra and Q2 is the part in the
second. Exactly how the Kahler reduction in structure group makes possible the
decomposition in (2.38) will be clarified in 3. These operators obey

Q1 2 = Q2 2 = {Q1 , Q2 } = 0, (2.39)

and the observables of Donaldson theory, such as O and I(), are annihilated by
both Q1 and Q2 .

In explicit computations via instantons, Q, Q1 , and Q2 correspond to the de


Rham exterior derivative d and to the and operators of instanton moduli space.

18
It is a familiar fact in cohomology of smooth, compact, Kahler manifolds that the
(complex-valued) de Rham cohomology is isomorphic to the cohomology. Some-
what analogously, in the Donaldson theory of Kahler manifolds, one can disregard
the Q2 symmetry and keep track of the topological behavior using Q1 alone; this
will be explained in 3.1.

Since Q1 lies in an N = 1 subalgebra, perturbations that break N = 2 super-


symmetry down to N = 1 can preserve Q1 invariance. Is there such a perturbation
that will simplify the problem?

2.5. The Mass

This question is no sooner asked than answered. Temporarily we work on a flat


R4 , with Euclidean coordinates y 1 , . . . , y 4 . Of course, we pick a complex structure,
determined say by complex coordinates z1 = y 1 + iy 2 , z2 = y 3 + iy 4 , to regard R4
as a Kahler manifold.

As was explained at the outset of this section, the N = 2 theory, regarded as


an N = 1 theory, has the following multiplets: there is a gauge multiplet (A, )
and a chiral supermultiplet = (, ) in the adjoint representation. The chiral
supermultiplet can have a bare mass
Z
L = m d4 xd2 Tr 2 h.c. (2.40)


preserving N = 1 supersymmetry. This can be written in the general form

X
a I(a ) + {Q1 , . . .}, (2.41)
a

so up to {Q1 , . . .}, by adding the mass term we are just shifting the Lagrangian by
a linear combination of the observables that we want to study anyway.

That is, it can be written this wayP in the case of a global Kahler manifold,
R as we will
see in 3. On flat R4 , the term a a I(a ) should better be written R4 Z, where
= dz1 dz2 . Globally, the two formulations are equivalent up to {Q1 , . . .}.

19
The N = 2 theory perturbed by the N = 1 invariant bare mass term (2.40)
has a mass gap for the following reasons. The perturbation certainly gives a mass
to the multiplet, so we are reduced at low energies to a minimal N = 1 system.
That system in turn has a dynamically generated mass gap, at a much lower scale
if m is large.

What are the global symmetries of the perturbed system? The minimal N = 1
model has a Z2h symmetry described earlier. The only symmetry of the multiplet
is , which generates a group that we will call Z2 . The overall global
symmetry group is hence Z2h Z2 . The pattern of symmetry breaking is easy
to work out. Z2h is broken to Z2 , as explained earlier, while the symmetry Z2
that acts only on fields with a positive bare mass (which can be taken large) is
unbroken. The unbroken symmetry group is therefore Z2 Z2 .

The Mass Term On A Curved Kahler Manifold

If we consider not flat R4 but a general Kahler manifold M, an important


subtlety arises in the description of the mass term. The subtlety arises because of
the twisting by U(1)R . The mass term in (2.40) is defined using a measure d4 x d2 ;
the factor d2 carries U(1)R charge (in the twisted model) and there is no natural
way to fix the space-time dependence of its phase on a curved Kahler manifold.

The problem can be remedied as follows. Write d4 xd2 = d2 z d2 zd2 . Then


the factor d2 zd2 is in fact naturally defined on any Kahler manifold (as should be
clearer, along with some subsequent claims, in 3), so the question is to generalize
d2 z.

The factor d2 z, or better m d2 z, including the factor of m from (2.40), can be


interpreted as a holomorphic two-form on R4 . The generalization is immediate:
on any Kahler manifold, pick a holomorphic two-form and replace m d2 z by .
The generalization of (2.40) to a general Kahler manifold is thus
Z
L = d2 zd2 Tr 2 + h.c. (2.42)
P
We will verify in 3 that this is of the form a a I(a ) + {Q1 , . . .}.

20
At this stage, however, an important restriction arises. A non-zero holomorphic
two-form exists not on arbitrary Kahler manifolds, but only on those on which
H 2,0 (M) 6= 0. So we are limited to Kahler manifolds obeying that condition.
This is in fact natural. One of Donaldsons earliest results [20] was a sort of
anomaly: the topological correlation functions of the twisted N = 2 theory are
in fact topological invariants only under a certain topological condition (the self-
dual part of H 2 (M) should have dimension bigger than one); specialized to Kahler
manifolds, the condition is precisely H 2,0 (M) 6= 0. So the restriction required by

our methods is in fact the right one.

2.6. Some Simple Computations

Now let us carry out some actual computations of the Donaldson invariants for
Kahler manifolds. Since there is a mass gap, we can proceed much as before; but

there are a few subtleties.

For simplicity we will specialize to gauge group G = SU(2). So the global


symmetry group is Z4 Z2 , spontaneously broken to Z2 Z2 ; as a result there
are two vacuum states, which we will call |+i and |i. The sum over vacuum states
in formulas such as (2.36) therefore involves two terms. The constants and hOi
that appear in that formula are odd under the broken symmetry, so + = and
hOi+ = hOi . We will normalize the operators O and I() so that = 1 and
hOi = 2. By comparing to special cases of Donaldson invariants that have been

There are two caveats here: (1) There is one other exceptional case of a Kahler manifold,
namely CP2 , for which the Donaldson invariants are well-defined; that is, the correlation
functions are independent of the metric. The construction of this paper does not apply
and I have no reason to expect the correlation functions to be given by similar formulas.
(2) Even when H 2,0 (M ) = 0 and the correlation functions of the twisted theory are not
topological invariants, they still have very nice properties (which have been much exploited
mathematically) and there may be a useful way to study them using physical techniques.
Apart from issues I discuss below, one must consider the violation of Lorentz invariance of
the twisted theory by the N = 1 invariant bare mass term. Lorentz invariance was used in
analyzing the two-forms to arrive at (2.36). We will show in 3 how to derive at (2.36), by
a slightly lengthier route, using properties that are preserved by the bare mass.

21
computed mathematically, one can verify that these normalizations agree with the
standard topological normalizations.

A subtlety arises in using the broken symmetry to relate the contributions of


the |+i and |i vacua in formulas such as (2.36). The global symmetry group
Z4 Z2 of the theory is a subgroup of the classical symmetry group consisting of
symmetries that are not explicitly broken by gauge instantons. However, gravita-
tional instantons in this case the effects of working on a four-manifold M might
induce additional symmetry breaking. We should remember that the Z4 symmetry
is generated by

: i, i. (2.43)

In an instanton field of instanton number k, the net number of minus zero


modes is, from the index theorem,

= 4k 3(1 h1,0 + h2,0 ), (2.44)

with hp,q the dimension of H p,q (M). This can also be written in a way that makes
sense for any four-manifold (but might not be integral in general):

3 1
= 4k ( + ) = dim M, (2.45)
4 2

with and the Euler characteristic and the signature of M, and M the moduli
space of instantons.

Under the symmetry (2.43), the integration measure for the fermion zero modes
transforms as i . Therefore, the underlying broken symmetry does not, as one
might expect, cause the constants C+ and C in formulas such as (2.36) to be
equal. Rather, the relation is

C = i C+ . (2.46)

Notice that the k-dependence drops out of this relation, because the spontaneously
broken symmetry has no anomaly under Yang-Mills instantons. Of course, similar

22
formulas can be worked out for groups other than SU(2) the difference being
that there are not two but h vacua, each with its own constant C.

Using (2.45), we can also write


 
3i
C = exp ( + ) C+ , (2.47)
8

showing that the relation between C and C+ is compatible with the fact that
each can be written as in (2.37).

Hyper-Kahler Manifolds

To work out actual formulas, we consider first the special case of hyper-Kahler
manifolds. This is the case in which the holomorphic two-form , which is respon-
sible for the mass gap of the theory, has no zeroes, so the mass gap can be applied
most simply.

Moreover, for a hyper-Kahler manifold, the holonomy is not SU(2)L U(1)R


(as for general Kahler manifolds) but simply SU(2)L . This means that the twisting
is trivial, and the physical model coincides with the topological model.

Actually, there are only two cases of compact, non-singular hyper-Kahler man-
ifolds of dimension four: a four-torus T4 , and a K3 surface. For T4 , h1,0 = 2,
h2,0 = 1, and = 0 modulo four, so C = C+ according to (2.46). In fact, for the
four-torus = = 0, and therefore C+ = C = 1, according to (2.37). Hence the
sum in (2.36) takes the following form in this case:

* !+
X 1X
exp a I(a ) + O = exp a b #(a b ) + 2
a
2
a,b

1 X
+ exp a b #(a b ) 2 .
2
a,b
(2.48)
Unfortunately, mathematical computations of Donaldson invariants of the four-
torus have not been performed, so there is nothing to compare (2.48) to. Inciden-

23
tally, it is straightforward to extend (2.48) to include the additional observables
associated with the odd dimensional cohomology of the torus.

It remains to consider the case of K3. For K3, h1,0 = 0, h2,0 = 1, so from
(2.46), C = C+ . Therefore, (2.48) is replaced by

* !+
X 1X
exp a I(a ) + O =C exp a b #(a b ) + 2
a
2
a,b

1X
exp a b #(a b ) 2 ,
2
a,b
(2.49)
with an unknown constant C. According to computations of [10], this formula is
in fact correct with C = 1/4.

Chiral symmetry breaking is the reason that the right hand sides of both (2.48)
and (2.49) are given by not a single exponential but a sum of two exponentials,
with a relative plus sign in one case and minus sign in the other case. These signs
play the following role. For the four-torus, for dimensional grounds explained in
(2.21), a correlation function

hO(x1 ) . . . O(xr )I(1 ) . . . I(s )i (2.50)

vanishes unless 4r + 2s is divisible by 8. For the K3 surface, the requirement is


that 4r + 2s should be congruent to 4 modulo 8. A single exponential of the form
in (2.48) or (2.49) would not obey these requirements, but the given sums of two
exponential with plus or minus signs do obey them. Thus (given the mass gap of
the N = 1 theory and the other constraints) chiral symmetry breaking is essential
in making sense out of Donaldson theory. This fact (of which we will see subtler
versions later, involving the canonical divisor) is a sort of global generalization of
old insights [1] concerning supersymmetric Yang-Mills theory on R4 .

24
2.7. General Kahler Manifolds

Now we consider the case of a general Kahler manifold M, with H 2,0 6= 0. The
important new phenomena spring from the fact that (if M is not hyper-Kahler)
the canonical class of M is non-trivial. This implies that the mass term

Z
L = d2 zd2 Tr 2 h.c. (2.51)
M

is not everywhere non-zero. The mass vanishes where does; and vanishes on
a divisor C representing the canonical class of M. For simplicity we will assume
at least initially that C is the union of smooth, disjoint Riemann surfaces Cy
along which has simple zeroes.

The local model that we reduce to by scaling up the metric of M is now the
following. Each Cy becomes a flat complex dimension one submanifold of a flat
R4 = C2 . One might think of the Cy as world-sheets of cosmic strings. The strings
in question are similar to global strings (their existence hinges on the vanishing
of a gauge-invariant field ) rather than local strings (in the core of which there
is an enhanced unbroken gauge symmetry).

The most important phenomenon associated with this particular type of cosmic
string is that it can capture fermion zero modes. In fact, pick complex coordinates
z1 , z2 on R4 , and suppose that = z1 dz1 dz2 , with a simple zero on the Riemann
surface C defined by z1 = 0. Any example (with non-singular Cy ) reduces locally
to this example when the metric is scaled up.

In this example, the mass term of the multiplet is proportional to z1 , so it


changes in phase by 2 in making a small circuit around the core of the string
at z1 = 0. This is the standard situation in which fermion zero modes are trapped
near the core of a string. In fact, the Dirac equation for the motion of the
field (the fermionic part of the multiplet) in the z1 plane has a normalizable
zero mode, as we will discuss in more detail in 3. When one allows for the z2

25
dependence of the wave-function, this zero mode becomes a two-dimensional
quantum field, propagating on the string world-sheet, that is, the z2 plane. We
will call this two dimensional quantum field . is a spin one-half fermion in
the adjoint representation of G (since those are the quantum numbers of ); it
has definite chirality, which we will conventionally call left-handed, since the zero
mode in the z1 plane has definite chirality.

In explaining the origin of the effective two-dimensional field , the gauge cou-
plings were not essential; the crucial zero mode develops because of the topology
of the Higgs field . When we consider the gauge fields, a new consideration
arises: the coupling of gauge fields to two-dimensional chiral fermions is anoma-
lous. Since the underlying four-dimensional theory is anomaly-free, there must be
additional fields trapped along the string that cancel the anomaly. To understand
the details, one would need to know something about how the strongly coupled
N = 1 supersymmetric Yang-Mills theory behaves near the core of the string.
This appears out of reach at present. But somehow, extra modes must be trapped
along the string in such a way that the effective theory along the string is a gauge-
invariant, anomaly-free theory that we will call the cosmic string theory. It can be
seen that the two-dimensional cosmic string theory has (0, 2) supersymmetry, but
we will not exploit that.

2.8. Symmetries And Vacuum Structure Of The Cosmic String The-


ory

Since we do not even know the fields and Lagrangian of the cosmic string
theory, how can we proceed? We will make what reasonable conjectures we can
based on the symmetries.

As we recall, the bulk theory had a Z2h Z2 global symmetry, which we


believe is spontaneously broken in bulk to Z2 Z2 . Along the string, the question
of symmetry breaking must be re-examined; further symmetry breaking might
occur in the core of the string.

26
In bulk, Z2 couples only to particles of positive bare mass, and it was im-
plausible that it would be spontaneously broken. Along the string, however, Z2
couples to (and in fact only to) the massless chiral fermion , so in the cosmic
string theory Z2 is in fact a chiral symmetry. It is very plausible that such a chiral
symmetry would be spontaneously broken.

On the other hand, a diagonal subgroup Z2 Z2 Z2 is generated by the


operator (1)F that counts fermions modulo two; this is equivalent to a 2 rotation
of the cosmic string, and cannot be spontaneously broken if rotation invariance is
unbroken.

We will make for the cosmic string theory the same sort of assumption that we
made in bulk. We will assume that the cosmic string theory has a mass gap and that
the only vacuum degeneracy comes from symmetry breaking. (Thus, in particular,
the vacuum degeneracy can be measured by local operators.) For the pattern of
symmetry breaking, I will assume that Z2 Z2 is spontaneously broken to Z2
the largest unbroken symmetry that permits fermion masses. Thus regardless of
the gauge group I postulate that the cosmic string theory has two vacuum states
(for a given choice of the vacuum in the bulk four-dimensional theory).

Hitherto we have considered the symmetries of the cosmic string theory which
are (i) not explicitly broken by Yang-Mills instantons; (ii) not spontaneously broken
in bulk. As in the bulk theory, for instance the discussion of (2.46), we must in
addition worry about whether there is further explicit breaking of symmetries by
gravitational instantons. What this means concretely is as follows. We consider
the actually relevant case in which the string world sheet is not the z2 plane but a
compact Riemann surface Cy ; we need to know whether the path integral measure
along Cy for the effective two-dimensional theory is even or odd under the broken
Z2 symmetry.

As in the bulk discussion, the question comes down to counting modulo two
the zero modes. If the number of zero modes is even, the measure is even under
Z2 ; if the number of zero modes is odd, the measure is odd. The number of zero

27
modes modulo two is a topological invariant, independent of the gauge fields (and
all fields other than ), so the problem can be treated as a problem of free fermions
along Cy .

Let d be the dimension of the gauge group. The field has d components; if
it is treated as a free field, the number of zero modes is d times what it would be
for a single chiral fermion. If d is even, there is no further issue; the number of
zero modes, and the path integral measure, are even. More subtle is the case of
an odd dimensional group, such as SU(2). The result then depends on the spin
structure. The number of zero modes is even or odd for chiral fermions coupled
to a so-called even or odd spin structure. Let y be 0 or 1 depending on whether
the spin structure is even or odd. Then the number of zero modes along Cy is
dy , modulo two, so the path integral measure transforms under the broken Z2
symmetry by a factor of

ty = (1)dy . (2.52)

The spin structure that arises here cannot be chosen arbitrarily but must be
deduced from the underlying four-dimensional theory. This will be done in 3.3
with the following result. If Cy M is a smooth Riemann surface defined by a
first order zero of a holomorphic two-form , then the inverse of the normal bundle
Ny to Cy in M can be interpreted as a spin bundle of Cy . This (or actually its
complex conjugate) is the spin bundle of which is a section; its evenness or
oddness determines (if d is odd) whether y is 0 or 1 and so whether the measure
is even or odd under the broken symmetry.

Computations

Now we come to the computational stage. Imitating our considerations in bulk,


the first step is to analyze the partition function of the cosmic string theory, with a
given choice of the bulk vacuum state |+i or |i and a given refinement to include
the vacuum structure along the string. We will label the four vacuum states in the
vicinity of a particular string as | + +i, | + i, | +i, and | i, with the first

28
sign indicating the ambient behavior in bulk and the second sign indicating the
behavior near the string.

As in the derivation of (2.24), the assumption of a mass gap implies that the
partition function of the cosmic string theory on a Riemann surface C must be of
the form

Z

h1iC = exp d2 x gW (x) (2.53)
C

with W (x) a local operator constructed from the Riemannian geometry. The only
topological invariant of a Riemann surface that can be constructed in this way
(indeed, its only topological invariant at all) is the Euler characteristic (C). It
would appear, therefore, that the partition function of the four-dimensional theory,
allowing for contributions from all of the Cy , would have a factor

!
Y X
exp (w(Cy )) = exp w (Cy ) , (2.54)
y y

with some universal coefficient w. However, using standard facts in complex ge-
P
ometry, y (Cy ) is a linear combination of (M) and (M), so this factor can
be absorbed in adjusting the constants a and b in (2.24).

We also have to consider the local operators that arise in the cosmic string
theory. The local operator O(x) of the original four-dimensional theory can be
inserted at points disjoint from the Cy , and so does not contribute to the cosmic
string theory. However, the other operator I() is constructed as an integral over
a two-manifold :
Z
I() = Zmn d mn . (2.55)

Topologically, intersections of with the Cy may be unavoidable; they can, how-


ever, be taken to be transverse.

29
We recall that when the metric of M is scaled up, the bulk contribution to
the integral in (2.55) can be discarded. In our earlier analysis, we obtained local
contributions at points of intersections of s. Similarly, local contributions will
arise at intersections of the a with the Cy . At every intersection point P of a and
Cy , the integral (2.55) will contribute a local operator V (P ) in the cosmic string
theory times a sign depending on the relative orientation of a and Cy . (As in
the discussion of intersections of s, such a sign must appear, since I() is odd
under reversal of orientation of .) So, if #( Cy ) is the algebraic intersection
number of and Cy , then one can make the replacement

X
I() #( Cy )Vy + terms involving intersections of s. (2.56)
y

Here Vy denotes the operator V inserted on Cy .

Inside correlation functions, the scalar operator V of the cosmic string theory
can be replaced by its vacuum expectation value like any local operator in a the-
ory with a mass gap. We must work out the expectation value of V in the various
vacuum states, which by our hypotheses are related by spontaneously broken sym-
metries. Let us therefore describe the symmetry and vacuum structure precisely.
Let and be the generators : i, and : , of the
Z4 and Z2 factors, respectively, of the global symmetry group Z4 Z2 . Then, as
we will see in 3.3,

V = iV, V = V. (2.57)

In bulk there are two vacua, |+i and |i, with

|+i = |i, |i = |+i. (2.58)

Allowing for the behavior near the string, these two states bifurcate into four, as

30
described earlier; we can choose the labeling of the vacua so that

| + +i = | +i
| +i = | + i
(2.59)
| + i = | i
| i = | + +i.

The action of need not be separately given as 2 is unbroken and hence acts
on the vacua as 2 . If therefore the expectation value of V in the state | + +i is v,
then altogether
hV i++ = v
hV i+ = iv
(2.60)
hV i+ = v
hV i = iv.

Final Evaluation

Finally, then, to obtain the topological correlation functions for Kahler man-
ifolds obeying our assumptions, one must sum over vacuum states in bulk and
along the Cy . Relative phases in the contributions of different vacua come from
(2.46) and (2.52). The parameters and ty appearing in those equations are not
independent, but are related by

X
+ y
= 0 modulo 2, (2.61)
y

(with ty = (1)y ), as we will explain in 3.4.

To compute correlation functions of I()s, one determines the cosmic string


contributions as in (2.56), and replaces V by its expectation values just given.

In writing down the final result, I will adjust the constants a and b in (2.24) and
the constant v in (2.60) to values chosen to agree with special cases of mathematical

31
computations. The requisite values turn out to be a = (7/4) ln 2, b = (11/4) ln 2,
and v = 1. The constants and hOi were similarly adjusted earlier, so we have
fixed five universal constants by comparing to particular calculations of Donaldson
invariants.

We want the general formula for

* !+
X
exp a I(a ) + O . (2.62)
a

It is convenient to set
X
y = a #(a Cy ). (2.63)
a

According to (2.56), the cosmic string contribution to the exponent in (2.62) is


P
y y Vy . Replacing this with its expectation values in the various states, we see
that the |+i vacuum, with its various bifurcations along the Cy , contributes


1 X a b Y 
2 4 (7+11) exp #(a b ) + 2 ey + ty ey . (2.64)
2 y
a,b

The |i vacuum, with its various bifurcations, contributes the chiral transform of
this, or


1 X a b Y 
(7+11) iy iy
i 2 4 exp #(a b ) 2 e + ty e . (2.65)
2 y
a,b

As a check, note that this is real by virtue of (2.61).

The final formula for the SU(2) Donaldson invariants of Kahler manifolds (with
H 2,0 (M) 6= 0 and possessing a holomorphic two-form whose zero set is a union of
smooth disjoint complex curves of multiplicity one) is obtained by summing these

32
contributions and is
* !+
X
exp a I(a ) + O
a T

1 X a b Y 
1+ 4 (7+11) y y
= 2 exp #(a b ) + 2 e + ty e
2 y
a,b

1 X a b Y 
+ i 21+ 4 (7+11) exp #(a b ) 2 eiy + ty eiy .
2 y
a,b
(2.66)
Here (as implied by the symbol h iT ) we use the normalization of topologists, and
so include a factor of 2 the order of the center of SU(2) from (2.18).

Here is a check on (2.66). Using (2.45) and (2.21), the dimension of the moduli
space of Yang-Mills instantons is 2, and the correlation function

hO(x1 ) . . . O(xr )I(1 ) . . . I(s )i (2.67)

is zero unless

4r + 2s = 2. (2.68)

Therefore, if we consider the a to be of degree 2 and to be of degree 4, then the


power series expansion of (2.66) should have only terms of degree congruent to 2
modulo 8. It is easy to verify this.

Unlike the naive formula (2.36) that would arise if the N = 2 theory had a
mass gap, the formula is not manifestly a topological (or differentiable) invariant.
Indeed, the formula appears to depend on the complex structure of M (which
determines the canonical divisor and thus the ultimately the y ). The metric
independence of the underlying twisted N = 2 theory ensures, however, that the
formula really only depends on the structure of M as a differentiable manifold. The
fact that the formula is a differentiable invariant but not obviously so is responsible
for its mathematical interest.

33
Thus, a fairly typical mathematical application of Donaldson theory is as fol-
lows. In many instances one can construct pairs of Kahler four-manifolds M1 and
M2 which have the same values of the obvious topological invariants but for which
the formulas (2.66) do not coincide. Then one can infer that M1 and M2 are in fact
not isomorphic as smooth manifolds. In practice, the data present in (2.66) beyond
the classical invariants are the cohomology classes of the connected components Cy
of the canonical divisor and the types (even or odd) of the normal bundles to the
Cy ; M1 and M2 must have the same values of this data if they are to coincide as
smooth four-manifolds.

Simple Type Condition And Comparison To Known Formulas


P
A four-manifold is said to be of simple type if f = hexp( a I(a ) + O)i
obeys the equation 2 f /2 = 4f . (2.66) evidently implies that Kahler manifolds
obeying our assumptions are of simple type. The derivation of (2.35) shows that
the simple type condition is really a corollary of having a mass gap. Thus, the
assumption that H 2,0 (M) 6= 0, which permitted us to reduce to a theory that has
a mass gap, was essential; we would not necessarily expect CP2 , for instance, to
obey a condition similar to the simple type condition.

In deriving (2.66) we assumed also that there exists a holomorphic two-form


that vanishes precisely on a union of disjoint smooth curves. This is not relevant to
the simple type condition, since regardless of the structure of the canonical divisor,
one can assume that O is inserted at points disjoint from it. Therefore, only the
bulk vacuum structure is relevant in determining the dependence of f on , so (at
a physical level of rigor) all Kahler manifolds with H 2,0 6= 0 are of simple type.

According to Kronheimer and Mrowka [11], on a four-manifold of simple type,


f , as a function of the a , is a Gaussian times a sum of exponentials. (In fact,
they make more precise claims.) This was a clue suggesting the role of a mass gap
and N = 1 supersymmetry, and (2.66) has the expected properties. We will say
more about this presently.

34
2.9. More General Kahler Manifolds

In general, we should consider the case of a Kahler manifold M with H 2,0 6= 0,


but with a canonical divisor that does not have the simplifying properties that we
have so far assumed. The zeroes of a holomorphic two-form will be a union of
connected curves Cy ; but even if is chosen generically the Cy may not be
smooth, and may vanish along components of Cy with multiplicity greater than
one.

No matter how badly behaved the canonical divisor may be, the mass gap
ensures that the contribution of a given Cy depends only on the local structure
of Cy and its intersections with the a . We recall that the Cy were defined as
the connected components of the zero locus of . Each Cy may in turn be a
union of irreducible (but perhaps singular) Riemann surfaces Cy, (which intersect
each other, perhaps at their singularities). The only homology invariants in the
intersections of the a and the Cy are the algebraic intersection numbers #(a
Cy, ), and the mass gap permits the correlation functions to depend on the natural
generalization of (2.63):
X
y, = a #(a Cy, ). (2.69)
a

The possible generalization of (2.66) allowed by the mass gap is


* !+
X
exp a I(a ) + O
a T

1 X a b Y
= 21+ 4 (7+11) exp #(a b ) + 2 Fy (y, )
2 y
a,b

1 X a b Y
+ i 21+ 4 (7+11) exp #(a b ) 2 Fy (iy, ),
2 y
a,b
(2.70)
where Fy (y, ) is a universal function, depending only on the local structure of
near Cy . Fy is a function of all the y, for Cy, a component of Cy .

35
We can be somewhat more precise about the nature of the function Fy . Let us
focus on a particular component Cy of the canonical divisor; we will simply call it
C, suppressing the y index. Let C , = 1 . . . w be the irreducible components of C,
P
and deleting the y index let = a a #(a C ). Each C has a multiplicity n
(the order of vanishing of the two-form in approaching a generic point of C ). The
vacuum state |+i of the bulk theory bifurcates along C to a number of vacuum
states |i i, where i runs over a finite set S . According to our assumptions
above, if n = 1, then S is a set with two elements. In what follows, I will make
no assumption about the number of vacua for n > 1, but I will assume that
the cosmic string theory has a mass gap and a finite vacuum degeneracy (which

can be detected by local operators) for any value of n . A vacuum state of the
cosmic string theory can be labeled by a w-plet ~ = (i1 , . . . , iw ). Indeed, once the
generic behavior along the C is given, all that remain are finitely many isolated
singular points; no further issues of infinite volume and vacuum degeneracy arise
in discussing the behavior near such isolated points.

The function F is simply the expectation in the cosmic string theory associ-
P
ated with C of the operator exp( a a I(a )). As in (2.56), the exponent can be
P
replaced by V where V is the operator V inserted at a generic point of
P
C and = a a #(a C ). In sum, we must calculate the expectation value
P
of the operator exp( V ) in the cosmic string theory. Scaling up the metric,
this reduces in the usual way to a sum over the contributions of vacuum states,

* +
X X
F = a~ exp V , (2.71)

~
~

In fact, a reasonable guess is that in general the cosmic string theory has 2n vacua. The
logic behind this is that a component C0 with a given multiplicity n0 arises from a two-
form with an n0 th order zero along a complex curve, say = z1 n0 dz1 dz2 . Such a
two-form
Qn0 can be deformed to a two-form with simple zeroes onn0n0 distinct curves, say
= i=1 (z1 ai )dz1 dz2 . After the perturbation, there are 2 vacua, coming from a
two-fold degeneracy on each of the n0 components. If the mass gap holds uniformly during
the process of deformation from to , then 2n0 is also the number of vacua prior to the
deformation.

36
with a~ being a constant measuring the amplitude with which the vacuum ~ ap-
pears. (Apart from the sort of effects considered in deriving (2.52) and (2.53),
the a~ might receive contributions involving the structure of the singularities of C.
The singularities, being isolated, do not affect the bulk or vacuum structure along
C, but might contribute volume-independent multiplicative factors.) This sum can
be factored as a product of contributions from the individual components C :

X w
Y 
F = a~ exp hV ii . (2.72)
~=(i1 ,...,iw )
=1

Here hV ii is the expectation value of V in the vacuum i . This is our final


result.

So in particular, F is a sum of exponential functions of the . In view of


how F enters in (2.70), this is in fact the general structure proved by Kronheimer
and Mrowka [11] for four-manifolds of simple type. We actually have somewhat
more precise information about the exponents that arise in F . (These exponents
are the simple classes in the language of Kronheimer and Mrowka.) For n = 1,
the expectation value hV ii equals 1 or 1, with our normalization above. For
general n , the possible values of hV ii depend only on n . The simple classes are
of the form
X
hV ii Cy, , (2.73)
y,

with i S and so are determined by the n .

If one is willing to make the assumption of the last footnote, one can be more
precise: the possible values of hV i are of the form 1 1 . . . 1 with a sum of
n terms, which are the contributions of the n ordinary cosmic strings into
which the multiplicity n string can be deformed. The expectation values are thus
n , n 2, n 4, . . . , n , and the simple classes are thus
X
ay, Cy, , (2.74)
y,

37
where the ay, are integers no greater in absolute value than the multiplicity ny,
and congruent to ny, modulo two. There is actually some additional evidence for
this formula: by considering the special case of an elliptic surface with a multiple
fiber, and comparing to a conjecture presented at the end of [11], one can see that if
their conjecture is true, the expectation values and simple classes are as just stated.
(The multiple fiber can have an arbitrary n , so a knowledge of the possible values
of hV i in this case determine the general structure of the simple classes.) It may
be that the list of simple classes is shorter than just claimed, as some of the a~
may vanish.

There remains the question of whether the constants a~ of equation (2.71)


contain any deep information about four-manifolds. I expect the answer to be
negative because the function F is invariant under a rather crude equivalence
relation. Because of the mass gap, in studying the cosmic string theory near C,
one can restrict to a small open neighborhood of U; because of the underlying
topological invariance, one can then deform the complex structure of U and the
two-form . I expect that using this relation one can eliminate the singularities of
C and reduce to the case that a connected component C of the canonical divisor is
a smooth curve of multiplicity 1 or a genus one curve of multiplicity n with normal
bundle of order n. In the former case, from our above results, F () = 2 cosh or
2 sinh depending on the type of the normal bundle; in the latter case, according
to the conjecture of Kronheimer and Mrowka, F () = sinh(n + 1)/ sinh .

Kahler Manifolds With 1 6= 0

To complete the discussion of Kahler manifolds, it remains to discuss the case


of a Kahler manifold M with 1 (M) 6= 0.

Actually, the fundamental group played no particular role in our considerations


(given that we assume that there are no unbroken and unconfined gauge symme-
tries), and as long as one restricts to the operators O(x) and I(), the correlation
functions are still given by the above formulas. The only real novelty is the follow-
ing. When the odd Betti numbers of M are non-zero, one can consider additional

38
R
operators I(T ) = T O(r) , with T an r-cycle in M for r = 1 or r = 3.

Correlation functions of the new operators can be analyzed with the same
methods used for I() and O. The mass gap means that the correlation functions
depend only on intersections of the various cycles with each other and with the
canonical divisor. Just as in our discussion of two-cycles, the intersections that
do not involve the canonical divisor detect only the classical intersection ring of
M. The interesting information comes from intersections involving the canonical
divisor C; the only such intersection involving odd-dimensional cycles is T T C
with T and T being three-cycles. Since = T T is a two-dimensional class,
we are again studying intersections C. The only novelty might be that the
operator V of the above discussion would be replaced by another operator of the
cosmic string theory, but I doubt that anything essentially new can be detected in
this way.

2.10. Other Gauge Groups

Before discussing other gauge groups in general, let us discuss one particularly
simple case: G = SO(3). Because G is locally isomorphic to G = SU(2), all local
considerations are unchanged. The symmetries, the symmetry breaking pattern,
the vacuum structure, and the constants a, b, , hOi, and hV i, are all determined by
the structure on R4 (perhaps in the presence of an infinite straight cosmic string)
and are unchanged in going from SU(2) to SO(3). Only a few global factors need
modification.

The only difference between the SO(3) theory and the SU(2) theory is that
there are SO(3) bundles that cannot be derived from SU(2) bundles. Given an
SO(3) bundle E, the obstruction to lifting its structure group to SU(2) is the sec-
ond Stiefel-Whitney class w2 (E). Before proceeding, we must formulate precisely
what we mean by the SO(3) theory. It is natural to define an SO(3) theory by
summing over all isomorphism classes of SO(3) bundles, but to get the sharpest
mathematical formulas, I will instead fix the second Stieffel-Whitney class and sum

39
over all bundles with a given value of w2 (E). Also, I will take the gauge group
to be the same as the SU(2) gauge group in the sense of only allowing gauge
transformations which, along any path on which the bundle E is trivialized, lift
to SU(2) gauge transformations. (Otherwise we would have to divide by an extra
factor equal to the number of components of the group of SO(3) gauge transforma-
tions; this is #H 1 (M, Z2 ). Also, as SO(3) has no center, the factor of 2 in (2.18)
in comparing different normalizations would not appear.)

The remaining global issues that entered the SU(2) analysis involved the action
of the global symmetries on the space of vacua. Since the SO(3) theory coincides
with the SU(2) theory on R4 , it has after perturbing by the bare mass the familiar
global symmetry group Z4 Z2 . However, on a four-manifold M, the path integral
measure may transform differently. As we recall, the transformation of the path
integral measure depends on index theorems that determine the number of fermion
zero modes both in bulk and along the cosmic strings.

If M is simply-connected, w2 (E) can be lifted to an integral cohomology class


(which we will also call w2 (E)) which is well-defined modulo two; hence w2 (E)2
is well-defined modulo four. Even if M is not simply-connected, one can still
make sense of w2 (E)2 modulo four, as noted in [7, p. 41]. The expression c2 (E) =
R 2
M Tr F F/8 (with Tr an invariant quadratic form on the Lie algebra normalized
to agree with the trace in the two-dimensional representation of SU(2)) is not
necessarily an integer if E cannot be derived from an SU(2) bundle. The relation
is rather
w2 (E) 2
c2 (E)
= modulo Z. (2.75)
4

Consequently, the formula (2.44) for the net number of zero modes is modified
to

E = 4k w2 (E)2 3(1 h1,0 + h2,0 ), (2.76)

with k Z. The path integral measure therefore transforms under the generator

40
of Z4 Z2 not as i but as iE , with

E = w2 (E)2 . (2.77)

Now let us consider what happens along a cosmic string component Cy . Let S
be the relevant spin bundle along Cy . The number of fermion zero modes modulo
two is a topological invariant of the bundle S E. If w2 (E) is zero when restricted
to Cy , then we can trivialize E along Cy ; then S E = S S S. The number
of zero modes of fermions coupled to S E is hence congruent modulo two to
the number y of zero modes of fermions coupled to S. Hence the transformation
law of the measure is ty,E = (1)y if (w2 (E), Cy ) = 0. On the other hand, if
(w2 (E), Cy ) 6= 0, then we can take E restricted to Cy to be O O(n) O(n),
where O(n) is any line bundle of odd degree n and O(n) is its inverse. So
SE = S S(n) S(n); for large positive n, the number of zero modes for
fermions coupled to those three line bundles are y , n, and 0. Since n is odd, this
is congruent to y + 1 modulo two, so the measure transforms as ty,E = (1)y +1
in this case. In either case, we can write

ty,E = (1)y +(w2 (E),C) . (2.78)

Now we can assemble the ingredients. Pick a class x H 2 (M, Z2 ), and let
h iT,x denote a topologically normalized correlation function, summed over all
isomorphism classes of SO(3) bundles E with w2 (E) = x. Then
* !+
X
exp a I(a ) + O
a T,x

1 X a b Y 
= 21+ 4 (7+11) exp #(a b ) + 2 ey + (1)(x,Cy ) ty ey
2 y
a,b

2 1 X a b Y 
+ ix 21+ 4 (7+11) exp #(a b ) 2 eiy + (1)(x,Cy ) ty eiy .
2 y
a,b
(2.79)
Thus, while the SU(2) theory determines the cohomology classes of the Cy modulo

41
torsion, the SO(3) theory determines the pairings of the Cy with arbitrary mod
two classes and thus is sensitive to the two-torsion. The additional information
is trivial for simply-connected M, but in general the SO(3) theory does contain
limited additional information.

That the right hand side of (2.79) is real follows from (2.61) together with the
following standard fact:

(x, x) + (x, C) = 0 modulo 2, (2.80)

P
with x any element of H 2 (M, Z2 ) and C = y Cy the canonical divisor (which
we are identifying with the Poincare dual cohomology class). Indeed, since C
reduces mod 2 to the second Stieffel-Whitney class w2 (M), this follows from the
Wu formula, which asserts that (x, x) = (x, w2 (M)) mod 2 for any x H 2 (M, Z2 ).
See [7, p. 6] for a quick proof of the Wu formula for the case that M is simply-
connected.

Groups Of Higher Rank

Now, let us briefly discuss what happens for gauge groups of higher rank.

One modification is that the global symmetry of the N = 2 theory perturbed


by a mass term that respects N = 1 invariance is not Z4 Z2 but Z2h Z2 ,
with h the dual Coxeter number. In bulk this is broken down to Z2 Z2 , so the
number of vacuum states in bulk is h instead of 2. Along the cosmic string it is
most plausible that Z2 Z2 is broken down to a diagonal Z2 subgroup (which
permits fermion masses), giving a two-fold degeneracy.

A more substantial change comes from the fact that for the gauge groups of rank
greater than one, there are higher Casimir operators. For instance, for G = SU(n),
(0) 1
(2.10) can be generalized to Or (x) = (2)r r! Tr r+1 (x) with 1 r n 1.
(k)
Applying the descent procedure, one gets operator valued k-forms Or (x), for
r = 1 . . . n 1. The main novelty is the occurrence of non-trivial operator-valued

42
(4)
four-forms (in contrast to O1 which is related to a classical invariant, the instanton
number). One can perturb the Lagrangian by

n1
X Z
(4)
LL+ br Or , (2.81)
r=2 M

with arbitrary coupling constants br ; this gives a family of topological field theories.

We can think of the br as infinitesimal variables in the sense that br nr = 0 for


some large, unspecified values of nr . This is sufficient for studying the correlation
functions, which is all we really want. Thus in particular, turning on the br does
not modify the mass gap and symmetry breaking pattern, which are stable against
(d)
small perturbations. After turning on the br , correlation functions of the Or with
d < 4 can be computed as we have done above for SU(2) in terms of a certain
number of universal quantities (in the SU(2) case they are , hOi, hV i, a, and b; for
higher rank there are more such constants because there are more operators). Once
the br are permitted to vary, the universal quantities are no longer constants but

become instead functions of the br .

Thus, a somewhat more elaborate structure arises in the case of Lie groups of
rank greater than one. However, the implications for topology of Kahler manifolds
may be limited, since one can only hope to detect the structure of C, but most of
the invariants of C that possess the necessary properties to be detected by theories
such as these ones are already detected by the SU(2) and SO(3) theories. Perhaps
with other groups one can learn more about the torsion part of the cohomology
classes of the Cy ; so far we have detected only the two-torsion (via the SO(3)
theory).

One can see such a phenomenon in the two-dimensional analog of Donaldson theory
where it occurs already for SU (2). The analog of the br are the i in eqn. (5.5) of [14];
the dependence on i arises through a change of variables given in eqn. (5.9) as a result of
which the constants that control correlation functions of Q(0) and Q(1) become functions
of the i .

43
3. Properties Of Supersymmetric Yang-Mills Theories

The purpose of this section is to fill in details suppressed in 2 concerning


the relevant properties of supersymmetric Yang-Mills theories. In doing so, we
will generally use conventions of Wess and Bagger [16]. For instance, doublets
of the SU(2)L (or SU(2)R ) rotation symmetries are represented by spinor indices
, . . . = 1, 2 (or , , . . . = 1, 2). Doublets of the internal SU(2)I symmetry will
be denoted by indices i, j, . . . = 1, 2. These indices are raised and lowered with the
antisymmetric tensor (or , ij ) with sign convention such that 12 = 1 = 21 .
Tangent vector indices to spacetime are denoted m, n, . . . = 1 . . . 4; the spinor and
tangent vector indices are related by the tensor m described in appendix A of
 
[16]; one similarly uses mn = m n m n /4, etc.

The fields of the minimal N = 2 supersymmetric Yang-Mills theory are the


following: a gauge field Am , fermions i and i transforming as (1/2, 0, 1/2) and
(0, 1/2, 1/2) under SU(2)L SU(2)R SU(2)I , and a complex scalar B all in
the adjoint representation of the gauge group. Covariant derivatives are defined by
Dm = (m + iAm ) and the Yang-Mills field strength is Fmn = m An n Am +
[Am , An ]. The supersymmetry generators transform as (1/2, 0, 1/2) (0, 1/2, 1/2);
introducing infinitesimal parameters i and i , they can be written

Am = i i m i + i i m i

i = mn i Fmn + i i D + i 2 m Dm Bij j

i = mn i Fmn i i D + i 2Dm Bij j (3.1)

B = 2 i i

B = 2 i i ,

with D = [B, B]. The minimal Lagrangian is (with signature + ++)


Z 
1 4 1
L = 2 d x Tr Fmn F mn i i m Dm i Dm BDm B
e 4
M (3.2)

1 2 i i j i ij
[B, B] Bij [ , ] + B [i , i ] .
2 2 2

44
These formulas have been adapted from the formulas of [16, p. 50] for N = 1
supersymmetry. Here Tr is an invariant quadratic form on the Lie algebra which
for G = SU(N) we can conveniently take to be the trace in the N dimensional
representation.

To construct a topological field theory, one replaces SU(2)R by a diagonal sub-


group of SU(2)R SU(2)I . This can be conveniently incorporated in the formalism
by replacing the internal indices i, j, . . . by another SU(2)R index . To describe
the topological transformation laws, one sets

i = 0,
(3.3)
=

(with an anticommuting parameter). This gives a one-component supersymmetry


which is the BRST symmetry of the twisted topological theory; the corresponding
charge will be called Q. The transformation laws are

Am = im

= i 2 m Dm B
= mn Fmn + i D (3.4)
B = 0

B = 2 .

The square of this transformation vanishes, up to a gauge transformation. By re-


quiring invariance under this topological symmetry, one can construct a topological
field theory on an arbitrary four-manifold M.

Now let us construct the observables of this theory. If we set

m = m
i (3.5)
B= ,
2 2

45
then the topological transformation laws become, in part,

Am = im
m = Dm (3.6)
= 0,

which is a standard topological field theory multiplet (related mathematically to


the equivariant cohomology of the gauge group acting on the space of connections).
The other fields do not contribute to the BRST cohomology.

The most obvious invariant operators are of the form O(x) = P ((x)), with
P an invariant polynomial on the Lie algebra of G and x a point in the four-
manifold M. For G = SU(2), the only essential choice of P is the quadratic
Casimir operator, and then we can take

1 1
O(x) = 2
Tr 2 = 2 Tr B 2 . (3.7)
8

Calling this operator O(0) (x) to stress that it can be regarded as a zero-form on
M, one next iteratively finds operator-valued k-forms O(k) , 1 k 4, obeying

dO(k) = {Q, O(k+1) }. (3.8)

For full details see [4]. In the case of a simply-connected four-manifold, the most
important of these operators is O(2) (as we explained in 2.2). The components of
O(2) were called Zmn in 2; explicitly

i  
Zmn = Tr 2 2BFmn + m n .
(3.9)
4 2

One now, as in 2, defines observables by


Z
I() = O(2) , (3.10)

with a two-dimensional homology cycle. From (3.8) one deduces that I() is

46
Q-invariant, and up to {Q, . . .} depends only on the homology class of . Equiva-
lently, one can introduce a closed two form and consider
Z
I() = O(2) . (3.11)
M

The two formulations are essentially equivalent, with being the Poincare dual of
.

3.1. Kahler Manifolds

Now we consider the further constructions that are possible if the metric on M
is Kahler. This means that the holonomy is not SU(2)L SU(2)R but SU(2)L
U(1)R , with U(1)R a subgroup of SU(2)R . The two-dimensional representation of
SU(2)R decomposes under U(1)R as a sum of two one-dimensional representations
(of charge 1/2). Accordingly, there are invariant projections of any doublet
v onto the components v 1 and v 2 transforming in definite U(1)R representations.
This permits one to define a complex structure on M; one simply declares that
the one-forms dxm m2 are of type (0, 1) while the one-forms dxm m1 are of type
(1, 0).

Similarly, one can decompose the topological generator of equation (3.3) into
two components, taking
i = 0
1 = 1 1 (3.12)
2 = 2 2

with anticommuting parameters 1 , 2 . The explicit 1 in (3.12) means that the 1


transformation is the part of the topological symmetry of (3.3) that comes from
an N = 1 subalgebra of the underlying N = 2 supersymmetry; likewise the 2
transformation comes from the second N = 1 subalgebra. Let Q1 , Q2 be the
charges corresponding to the 1 and 2 transformations. Then Q = Q1 + Q2 , and
Q1 2 = Q2 2 = {Q1 , Q2 } = 0.

47
The vertex operators O(k) are invariant under both the 1 and 2 transforma-
tions, modulo the equations of motion. The fact that the equations of motion enter
here will be important presently.

In this paper, it will suffice to consider the 1 symmetry, with 2 = 0. The


transformation laws are

Am = i1 m1 1

1 = i 21 m 1 Dm B
2 = 0

1 = 1 mn 1 Fmn i1 D (3.13)

2 = 0
B = 0

B = 21 12 .

Certain of these equations have the following interpretation. If we decompose


F into its parts F 2,0 , F 1,1 , and F 0,2 of the indicated types, then 11 is proportional
to F 0,2 , so a BRST-invariant configuration must have F 0,2 = 0; hence it describes
a holomorphic vector bundle. The first equation shows that the (0, 1) part of the
connection A is BRST-invariant; hence the holomorphic structure of the bundle is
BRST-invariant.

To justify a claim made at the end of 2.4, the reason that Q1 -invariance is
enough in analyzing the topological correlation functions is that the change in
the Lagrangian under a change in the Kahler metric or coupling constant can be
written in the form {Q1 , . . .}. We can see this as follows. Since Q1 lies in an N = 1
algebra, it is convenient to recall the standard way of writing the N = 2 theory
R
in N = 1 superspace. The part involving the multiplet is d4 xd2 d2 Tr .
R
Since Q1 acts (up to a total derivative) as d for suitable , this is {Q1 , . . .},
and its variation with respect to metric or coupling has the same property. The
R R 2
gauge kinetic energy is d4 xd2 Tr W 2 + d4 xd2 Tr W . The first term here is

48
{Q1 , . . .} for the same reason as before, while (by a standard identity) the second
coincides with the first up to a multiple of the instanton number. Hence a change
in Kahler metric or gauge coupling induces at most a term of the form {Q, . . .}
plus a change in the angle (which is the coupling that multiplies the instanton
number). Using the familiar chiral anomaly of the U(1) R-current, a change in the
angle can be absorbed in a chiral rotation or rescaling of the fields I() and O
and (allowing for the gravitational part of the chiral anomaly) a redefinition of the
constants called a and b in (2.24). The rescaling of I() and O merely affect the
values of the constants , hOi, and v that entered our formulas.

The Mass Term

Exploiting the fact that Q1 is the only essential symmetry, we want to introduce
a mass term for some of the fields, preserving the Q1 symmetry. As was foreseen
in 2, this construction depends on a choice of holomorphic two-form on M.
R
We consider adding to the Lagrangian I() = M O(2) . This contains a term
R
M Tr BF . Since is of type (2, 0), only the (0, 2) part of F enters here; as
noted after equation (3.13), this is of the form {Q1 , . . .}. So modulo {Q1 , . . .}, the
BF term can be dropped and I() is proportional to
Z

L1 = 1/2 Tr 2 2 mn22 kl mnkl gd4 x. (3.14)
M

Now, in verifying the Q1 invariance of L1 (or equivalently, of I()) it is neces-


sary to use the equations of motion. In fact, a small computation shows that
Z

L1 = i 21 Tr B2 m Dm 2 np22 kl npkl gd4 x. (3.15)
M

This vanishes by the equation of motion of , as one can see by considering the
part of the original Lagrangian L that contains . This is
Z  
i 4 m 2
2
L0 = 2 d x g Tr 2 Dm + 22 [B, ] . (3.16)
e
Instead of saying that L1 is Q1 invariant modulo the equations of motion derived

49
from L, one can modify the Q1 transformation laws so that the full Lagrangian
L + L1 is Q1 -invariant. The requisite correction to the transformation laws is that,
instead of vanishing, 2 should be as follows:


2 = 2e2 1 B2 np22 kl npkl . (3.17)

Since B = 0, this correction to the transformation law does not affect the fact
that Q1 2 = 0. Nor does it affect the Q1 invariance of the various observables of
the theory (such as O or I()), as they are independent of .

While preserving Q1 invariance, we could add to the Lagrangian any expression


of the form {Q1 , V }. It is convenient to do so with the particular choice
Z
1
V = d4 x g Tr B 2 2 np kl npkl . (3.18)
2

The corresponding addition to the Lagrangian is


Z Z 2
1
L2 = 4
d x g 2 2 np 22 kl npkl e2 d4 x g Tr BB np22 kl npkl .
2
(3.19)

So we consider the combined Lagrangian

b =L + L1 + L2 = L + I() + {Q1 , . . .}
L
Z   Z
1 4 2 2 2 (3.20)
=L d x Tr m + m 2 2 e d4 x 2mm Tr BB.
2
M M

with

m = mn22 kl mnkl . (3.21)

As promised in 2, we have realized a mass term for the N = 1 matter multiplet


(which consists of B and = 2 ) by adding to the Lagrangian a term of the
form I() + {Q1 , . . .}. The mass is proportional to the holomorphic two-form ,
as expected. The N = 1 gauge multiplet, consisting of the gauge field Am and the
gluino = 1 , remains massless.

50
Symmetries

For subsequent use, it will be important to determine the transformation law


of the observables O and I() under the global symmetries of the theory. (For a
related analysis, see [13].)

As we discussed in 2, the twisted theory, perturbed by the mass term as in


(3.20), has a global symmetry Z4 Z2 . We will describe the symmetries in the
language of N = 1 supersymmetry, the N = 1 multiplets being (Am , ) and (B, ).
The generator of Z4 acts by

() = i, () = i, (B) = iB, (3.22)

with Am , being invariant. (Such a symmetry, which acts differently on different


components of a supermultiplet, is usually called an R symmetry.) The non-trivial
element of Z2 acts by

(B) = B, () = , (3.23)

with the gauge multiplet (Am , ) being invariant.

Now let us determine the transformation laws of the observables. O(x) =


Tr B 2 / 2 evidently transforms as (1, 1) under and . Now let us consider the
quantum numbers of I() with a (p, q)-form (with p+q = 2); we will schematically
R
call this operator I p,q . I 1,1 is of the general form M ( + BF ); therefore,
from the transformation laws presented in the last paragraph, I 1,1 transforms as
(i, 1).

What about I 2,0 ? This is analyzed most efficiently by noting that I 2,0 is
present in the Lagrangian, so it must be invariant, transforming as (1, 1), modulo
{Q1 , . . .}. If one tries to verify this explicitly, one has schematically I 2,0
= +
BF in the original twisted topological field theory. The two terms transform
differently under the symmetries, but the second is {Q1 , . . .}, a fact exploited

51
above, and so can be dropped. It appears that we can take I 2,0
= , which
is invariant, as expected. This is the correct structure of I 2,0 in the absence of
the mass perturbation introduced above, but that perturbation brings about the
following change. We recall that I 2,0 was only Q1 -invariant modulo the equations
of motion. In adding a perturbation L2 = {Q1 , V } to the Lagrangian to complete
the mass term (3.20), we have disturbed the equations of motion, and accordingly
an extra term must be added to I 2,0 . The result can be found most simply by noting
b
that (before or after addition of L2 ) I 2,0 can be interpreted as L/m, with m as
in (3.21). This gives an extra term L2 /m, and finally I 2,0 = + BB.
R
Finally, we consider I 0,2 , which is of the general form M (+BF ). Again,
the two terms transform differently under the symmetries. The second term can

be written as {Q1 , . . .} using the correction (3.17) to the transformation laws, and
so is again inessential. So I 0,2 transforms as (1, 1).

To summarize this discussion, the quantum numbers of the observables are as


follows:
O : (1, 1)
I 2,0 : (1, 1)
(3.24)
I 1,1 : (i, 1)
I 0,2 : (1, 1).

Moreover modulo {Q1 , . . .}, we have schematically

I 1,1
= + BF
(3.25)
I 0,2
= .

Gluino Condensation

Let us now analyze in bulk the dynamical effects of the mass perturbation in
(3.20). This means that we work on a flat R4 and take = dz1 dz2 ; of course

This is actually only true for = , but that is good enough for probing Donaldson theory
as was in any case an arbitrary holomorphic two-form.

52
any example, suitably scaled up, looks like this locally as long as one keeps away
from the zeroes of .

Seen as an N = 1 theory, the N = 2 supersymmetric Yang-Mills theory has


a gauge multiplet (Am , ) and a matter multiplet = (B, ). The perturbation
in (3.20) is in bulk simply the usual N = 1 invariant bare mass term for the
multiplet. Its addition to the Lagrangian leaves at low energies the pure N = 1 su-
persymmetric gauge theory of (Am , ). This theory has a Z4 symmetry, generated
by the transformation described in (3.22). It is believed that this symmetry is
broken to its Z2 subgroup, by gluino condensation, that is, by the expectation
value of Tr , which is essentially our friend I 0,2 . We will abbreviate Tr
as .

It will be necessary therefore for us to discuss some standard properties of


gluino condensation. Let us recall the basic structure of the renormalization group.
If a quantum theory is defined using a renormalization point and with a coupling
constant g, then the effective coupling constant g obeys

 

+ (g) g = 0, (3.26)
g

with (g) the beta function. If g is defined by measurements at an energy E,


then g is a function only of the dimensionless variables E/ and g. If for weak
coupling the beta function looks like = bg 3 + . . . (with b > 0 for asymptotically
free theories) then the renormalization group equation can be written in the weak
coupling regime as

 
2 2
2b(g ) g 2 (E/, g) = 0. (3.27)
g 2

The solution is
1 1
= 2b ln(/E). (3.28)
g 2 (E/) g2

53
Strong coupling arises for g
= 1 or

E
= exp(1/2bg 2 ). (3.29)

Even though a weak coupling approximation was used in solving (3.26), (3.29)
gives correctly the singular behavior of E for small g which is in fact dominated
by the behavior in the weakly coupled regime.

Now, we are interested in the case in which at some mass scale m, an N = 2


theory is explicitly broken down to an N = 1 theory. The N = 2 theory has a beta
function = bg 3 + . . ., and the N = 1 theory has a beta function = b g 3 + . . ..
The relation between them is
2b
b= . (3.30)
3
(In fact, the beta function of supersymmetric Yang-Mills theory of general N is
proportional to N 4 and has a well-known zero for N = 4.) We suppose that
m << and is large enough to be in the weak coupling regime. In that case, as
long as E m, we can use the solution (3.28) for the effective coupling. However,
for E m, we must evolve the effective coupling using the beta function of the
low energy theory. In doing this, we use (3.28) to determine the initial conditions
for the renormalization group evolution of the low energy theory. In fact, for
the renormalization scale L and coupling gL of the low energy theory, we take
the values L = m and gL = g(m/, g) that are appropriate for the microscopic
N = 2 theory at the scale m. The subsequent evolution of the effective coupling
to energies below m is carried out using an equation just like (3.26), with , g, g,
and replaced by m, g(m/, g), the effective coupling g L of the low energy theory,
and the beta function of the low energy theory. So we get the analog of (3.28):
1 1
2 = 2 2b ln(m/E). (3.31)
g L 2 (E/m, g (m/, g)) g (m/, g)

These equations combine to give


1 1
= 2b ln(/m) 2b ln(m/E). (3.32)
gL2 g2

54
Gluino condensation occurs for g L of order one, which occurs for

E

= exp(1/2b g 2 )m1(b/b ) b/b . (3.33)

In our particular case, using (3.30), the dependence of the energy scale on m
and is

E
= m1/3 2/3 . (3.34)

Since the gluino condensate has dimension three and so scales as E 3 , we get

hi
= m2 , (3.35)

a result that will be essential later. This estimate of the m dependence of hi is


reliable for m in the weak coupling regime a harmless restriction in the topological
theory as the coupling constant dependence of the Lagrangian is {Q, . . .}. It is
possible to show that the m dependence given in (3.35) is exact using arguments
of holomorphicity, as in [2,21].

Of course, in (3.35) we have determined only the m dependence of the gluino


condensate, not the overall constant multiplying this. Because of the basic struc-
ture of Z4 broken to Z2 , there are really two vacua with hi
= m2 , times a
constant independent of m and .

3.2. Evaluation Of Correlation Functions In Bulk

Apart from gluino condensation, the N = 1 theory is also believed to have a


mass gap. In 2.3, we sketched how the topological correlation functions are deter-
mined by the mass gap. I will here fill in various details omitted there, involving
the quantum numbers of the operators, the precise role of gluino condensation, and
the lack of Lorentz invariance (even in bulk) of the mass term that breaks N = 2
down to N = 1. The reason for the latter statement is that, in the twisted theory,
the mass term comes from a choice of holomorphic two-form (say = dz1 dz2 on
a flat R4 ) which breaks the SO(4) rotation symmetry of R4 down to SU(2).

55
First we consider correlation functions of the operator O = Tr B 2 / 2 . This
operator transforms as (1, 1), so it can get an expectation value once the global
symmetry Z4 Z2 is spontaneously broken to Z2 Z2 ; this expectation value
will, of course, be odd under the broken symmetry. To see how this comes about
more explicitly, we integrate out the massive (B, ) multiplet to express O purely
in terms of light fields. This can be done via the one loop Feynman diagram of
figure (1), with the result that O can be replaced by /m, times a constant that

I will not compute here. On the other hand, according to (3.35), the expectation
value of scales as m2 . So we get

hOi = 2 C, (3.36)

with the sign depending on the sign of the gluino condensate and C a universal
constant; 2 C is the universal constant hOi of 2.

Two-Forms

As a prelude to incorporating the two-form operators, let us compare correla-


tion functions h i of topological observables in the twisted theory to correlation
functions h i1 in the theory perturbed by a mass term. The relation is

hA1 A2 . . . An i1 = hA1 A2 . . . An eI() i, (3.37)

since the mass term was I() + {Q1 . . .}.

The main difference between the present problem and the case considered in
2.3 is the lack of Lorentz invariance in the bulk theory, once a mass term is
introduced via a choice of holomorphic two-form . In 2.3, Lorentz invariance
was used to prove that the vacuum expectation value of I() or equivalently I()
vanishes. This will no longer be true.

The factor of 1/m is needed on dimensional grounds: O and have dimension 2 and
3, respectively. By examining the diagram, it is easy to see that one gets 1/m and not
1/m1a ma for a 6= 0; this actually follows from general principles of holomorphicity.

56
If we look at the quantum numbers in (3.24), and take into account the fact
that Z4 Z2 is spontaneously broken down to Z2 Z2 , it is clear that I 1,1 cannot
have a vacuum expectation value, and it appears that either I 0,2 or I 2,0 might have
expectation values.

In fact, the unbroken supersymmetry of the N = 1 theory implies that h1i1 = 0


in bulk and hence that

hexp I()i = 0 (3.38)

for a holomorphic two-form. Thus, I 2,0 has no vacuum expectation value.

However, according to (3.25), I 0,2 is proportional to our friend the gluino bi-
linear and thus has a vacuum expectation value due to gluino condensation.
According to (3.35), the expectation value of is proportional to the mass m or
R
in other words to the two-form . The operator I 0,2 () = M Z is of course
also proportional to the two-form . So one gets in fact
Z
hI()i1 = 0 , (3.39)
M

with 0 a universal constant that is proportional to the gluino condensate and


therefore odd under the spontaneously broken symmetry. This is equivalent to
Z
I()
hI()e i = 0 . (3.40)
M

A consequence of (3.39) is as follows. If is any Riemann surface in M (not


necessarily holomorphically embedded) then
Z
hI()i1 = 0 . (3.41)

This arises upon expanding the Poincare dual of in classes of type (2, 0), (1, 1),
and (0, 2), recalling that I p,q has an expectation value only for (p, q) = (0, 2), and
using (3.39).

57
Now let us consider the general case of several two-forms. We still, however,
temporarily keep away from zeros of the holomorphic two-form ; those will be
considered later.

As in 2.3, to exploit the mass gap it is convenient to introduce s Riemann


surfaces 1 , . . . , s in space-time, which we can assume to intersect only pairwise.
(The a are not necessarily holomorphically embedded.) As we saw in 2.3, us-
ing Lorentz invariance of the bulk theory and assuming that the mass gap holds
everywhere, one gets

* !+
X X
exp a I(a ) = exp a b #I(a b ) . (3.42)
a 1 a,b

with some universal constant . We recall that in the derivation of (3.42), one uses
the fact that (by Lorentz invariance) the expectation value of I(a ) vanishes in
bulk. In the present case, that is no longer so; the expectation values are given in
(3.41). Including the appropriate extra contributions, we get

* !+
X X X Z
exp a I(a ) = exp a b #I(a b ) + 0 a ,
a 1 a,b a
a
(3.43)
just as if I() is a free or Gaussian field with a one point function given in (3.41)
and a two point fuction given in (2.33).

This is equivalent to

* !+
X X X Z
exp a I(a ) + I() = exp a b #I(a b ) + 0 a .
a a,b a
a
(3.44)
It must be the case that 0 = , since otherwise the formula is not invariant under
a change in the unnatural splitting of the exponent on the left hand side between

58
I() and the I(a ). For 0 = , we can reexpress (3.44) in a more invariant form
by expanding I() in terms of the I(a ) (plus {Q1 , . . .}):

* !+
X X
exp a I(a ) = exp a b #I(a b ) . (3.45)
a a,b

Thus, we have recovered the key formula (2.35) of 2.3, though in a more round-
about way because of the lack of manifest Lorentz invariance in the N = 1 formu-
lation.

Before leaving this subject, I will comment briefly on the computation of .


Near intersections of s, one can assume that locally the s look like holomor-
phically embedded Riemann surfaces. For instance, locally 1 could be the locus
R
z1 = 0 and 2 the locus z2 = 0 in R4 = C2 . In this case, I(1 ) = z1 =0 I 1,1 and
similarly for I(2 ). Since I 1,1
= Tr( + BF ), we have to evaluate something like
Z
Tr( + BF )(z1 ) Tr( + BF )(z2 ). (3.46)
1 2

The role of the mass gap is here very explicit since B, have bare masses in
the N = 1 theory. Integrating out these massive fields, the integral in (3.46)
can be expressed in terms of operators of the low energy theory that is, the
N = 1 theory of the gauge multiplet Am , . The leading contribution comes from
integrating out the field in (3.46) and is simply the local operator /m evaluated
at z1 = z2 = 0. Since the gluino condensate is of the general form m with a
constant independent of m, the factors of m cancel out and the local contribution
to the expectation value of (3.46) is independent of m, as expected. Of course, this
contribution is odd under the broken symmetry, as promised in 2; the underlying
reason for this is that the operator I 1,1 transforms as (i, 1) according to (3.24).

59
3.3. Behavior Near Cosmic Strings

Now we want to analyze the behavior near a cosmic string. We work on flat R4
with Euclidean coordinates y 1 . . . y 4 and complex coordinates z1 = y 1 + iy 2 , z2 =
y 3 + iy 4 . With the conventions of Wess and Bagger [16] (rotated from Minkowski
space to Euclidean space by setting y 0 = iy 4 ), the Dirac operator is

D D
!
Dz2 Dz1
m Dm = 2 D D
. (3.47)
Dz 1 Dz 2

We consider the cosmic string defined by the holomorphic two-form = z1 dz1 dz2 ,
which vanishes on the curve C defined by z1 = 0. Away from C, the field = 2
has a bare mass. This mass vanishes in the core of the string.

As we claimed in 2.7, there is a normalizable zero mode for the field in the
1
z1 plane. In fact, the equations of motion of 1 and , allowing for the kinetic
energy and the coupling to , are

1 1
i z1 = 0
z1
1
(3.48)

i + z1 1 = 0.
z 1

1 2
The zero mode is 1 = exp(|z1 |2 ), = i exp(|z1 |2 ), with 2 = = 0.

Now we wish to consider the motion in the z2 direction. To do so, we write


1
1 = exp(|z1 |2 ) (z2 , z 2 ), = i exp(|z1 |2 ) (z2 , z 2 ), with a two-dimensional
fermi field, defined on C, and with values in the adjoint representation. With this
ansatz, the kinetic energy of the field reduces to
Z
D
dz2 dz 2 Tr . (3.49)
Dz2
C

This shows that, as claimed in 2.7, behaves like a chiral fermi field along C.

60
To determine the spin structure to which is coupled, we consider a global
situation in which C is a Riemann surface in a four-manifold M defined by the
vanishing of a holomorphic two-form . Locally one can write = z1 dz1 dz2 ,
with z1 and z2 being local coordinates normal and tangent to C. The existence of
a global two-form of the stated behavior implies that dz1 transforms as (dz2 )1/2 .
To say this in a more sophisticated way, the normal bundle NC to C obeys NC 2 =
KC 1 , with KC the canonical bundle to C. Thus a section of NC can be interpreted
as a (1/2, 0) form on C. Hence NC 1 whose sections transform as /z1 is
a spin bundle of C, that is, a bundle of (1/2, 0) forms. The complex conjugate
bundle N C 1 is a bundle of (0, 1/2) forms or an antiholomorphic spin bundle.

On the other hand, the formula (3.13) shows that transforms as DB/Dz 1 ,
to which it is related by the symmetry. Thus, transforms as a section of the
antiholomorphic spin bundle N C 1 , as was claimed in 2.8.

Quantum Numbers Of V

The last claim made about cosmic strings in 2 was that the operator V has
the quantum numbers claimed in (2.57).

In fact, as in equation (2.56), V is the contribution to integrals such as I() =


R
M Z coming from the intersection with a cosmic string. As the Poincare dual
of the cosmic string is a class of type (1, 1), only the (1, 1) part of contributes;
hence V transforms under the symmetry as does I 1,1 , and (2.57) is a consequence
of (3.24).

3.4. A Useful Formula

Finally, we must establish the formula (2.61) that was needed to show that the
final expression for the Donaldson invariants is real.

We let be a holomorphic two-form on M that vanishes on a divisor C. Let


K be the canonical bundle of M and K|C its restriction to C. There is an exact

61
sequence of sheaves

0 OK K|C 0. (3.50)

The first map is f f . This leads to an exact sequence

1 1 (M, K) . . .
0 (K) H 0 (C, K| ) H 1 (M, O)H
0 H 0 (M, O)H C

(3.51)
We truncate this to

0 (K) H 0 (C, K| ) ker( 1) 0.


0 H 0 (M, O)H (3.52)
C

The existence of such an exact sequence implies that

dim H 0 (M, O) dim H 0 (M, K) + dim H 0 (C, K|C ) dim ker( 1 ) = 0. (3.53)

The map 1 can be interpreted as the map from H 1 (M, O)


= H 0,1 (M, C)
to H 1 (M, K)
= H 2,1 (M, C). A natural antisymmetric bilinear form on
H 0,1 (M) can be defined by

Z
h, i = . (3.54)
M

By Poincare duality, vanishes as an element of H 2,1 (M, C) if and only if


h, i = 0 for all .

Now in general, if V is a vector space of finite dimension and h , i is an


antisymmetric bilinear form on V , then the kernel of h , i is defined as the set
of x V such that hx, yi = 0 for all y V ; simple linear algebra shows that the
dimension of the kernel is congruent to the dimension of V modulo two. The facts
stated in the last paragraph mean that ker 1 is the same as the kernel of h , i

62
and hence that

dim ker 1
= dim H 0,1 (M) mod 2. (3.55)

This lets us rewrite (3.53) in the form

dim H 0 (C, K|C )


= 1 h0,1 (M) + h0,2 (M) mod 2. (3.56)

(We recall that hp,q (M) = dimH p,q (M).)

On the other hand, if C is the union of connected components Cy , then


H 0 (C, K|C ) = y H 0 (Cy , K|Cy ). Above we showed in essence that K|Cy is the
spin bundle of the fermions that propagate along Cy (actually, the chiral fermions
are coupled to the antiholomorphic spin bundle that is complex conjugate to
K|Cy ). So the number of fermion zero modes along Cy is y = dim H 0 (Cy , K|Cy ).
Taking account of the formula for in (2.44), it follows then that (3.56) is equiv-
alent to
X

= y mod 2. (3.57)
y

This is the desired formula used in 2.

63
FIGURE CAPTIONS

1) A one loop diagram by which the operator O can be expressed as an effective


operator in the low energy theory.

64
REFERENCES

1. V. Novikov, M. Shifman, A. Vainshtein, and V. Zakharov, Nucl. Phys. B229


(1983) 381, 407.

2. D. Amati, K. Konishi, Y. Meurice, G. C. Rossi, and G. Veneziano, Non-


Perturbative Aspects in Supersymmetric Gauge Theories, Phys. Rep. 162
(1988) 169.

3. I. Affleck, M. Dine, and N. Seiberg, Nucl. Phys. B241 (1984) 493.

4. E. Witten, Topological Quantum Field Theory, Comm. Math. Phys. 117


(1988) 353.

5. L. Baulieu and I. M. Singer, Topological Yang-Mills Symmetry, Nucl.


Phys. (Proc. Suppl.) 15B (1988) 12.

6. S. Donaldson, Polynomial Invariants For Smooth Four-Manifolds, Topol-


ogy 29 (1990) 257.

7. S. Donaldson and P. Kronheimer, The Geometry Of Four-Manifolds,


Oxford University Press (1990).

8. R. Friedman and J. W. Morgan, Algebraic Surfaces And Four-Manifolds:


Some Conjectures and Speculations, Bull. Am. Math. Soc. 18 (1988) 1,
Smooth Four-Manifolds And Complex Surfaces, to appear.

9. S. Donaldson, Gauge Theory And Four-Manifold Topology, preprint.

10. K. G. OGrady, Donaldsons Polynomials For K3 Surfaces, J. of Diff.


Geom. 35 (1992) 415.

11. P. Kronheimer and T. Mrowka, Recurrence Relations And Asymptotics For


Four-Manifold Invariants, submitted to Bull. Amer. Math. Soc.

12. E. Witten, The Verlinde Algebra And The Cohomology Of The Grassman-
nian, IAS preprint HEP-93/41.

65
13. J.-S. Park, Holomorphic Yang-Mills Theory On Compact Kahler Mani-
folds, N = 2 Topological Yang-Mills Theory On Compact Kahler Sur-
faces, ESANAT and Yonsei preprints (1993).

14. E. Witten, Two Dimensional Gauge Theories Revisited, J. Geom. Phys. 9


(1992) 303.

15. A. Johansen, Twisting Of N = 1 SUSY Gauge Theories and Heterotic


Topological Models, Fermilab preprint.

16. J. Wess and J. Bagger, Supersymmetry And Supergravity (Princeton Univer-


sity Press, 1983).

17. N. Seiberg, Supersymmetry And Non-perturbative Beta Functions, Nucl.


Phys. 208B (1988) 75.

18. E. Witten, On Quantum Gauge Theories In Two Dimensions, Commun.


Math. Phys. 141 (1991) 153.

19. R. Gompf and T. Mrowka, A Family Of Non-complex Homotopy K3


Surfaces, to appear in Annals of Math.

20. S. Donaldson, Irrationality And The h-Cobordism Conjecture, Jour. Diff.


Geom. 26 (1987) 141.

21. N. Seiberg, Naturalness Versus Supersymmetric Nonrenormalization Theo-


rems, Phys. Lett. 318B (1993) 469.

66
This figure "fig1-1.png" is available in "png" format from:

http://arXiv.org/ps/hep-th/9403195v1

You might also like