hep-th/9606040

Mirror Symmetry is T-Duality

arXiv:hep-th/9606040v2 14 Jun 1996

Andrew Strominger†

Department of Physics, University of California, Santa Barbara, CA 93106, USA

Shing-Tung Yau†† and Eric Zaslow†††

Department of Mathematics, Harvard University, Cambridge, MA 02138, USA

Abstract

It is argued that every Calabi-Yau manifold X with a mirror Y admits a family of

supersymmetric toroidal 3-cycles. Moreover the moduli space of such cycles together with
their flat connections is precisely the space Y . The mirror transformation is equivalent
to T-duality on the 3-cycles. The geometry of moduli space is addressed in a general
framework. Several examples are discussed.

†
email: andy@denali.physics.ucsb.edu
††
email: yau@abel.math.harvard.edu
†††
email: zaslow@abel.math.harvard.edu
1. Introduction

The discovery of mirror symmetry in string theory [1] has led to a number of mathe-
matical surprises. Most investigations have focused on the implications of mirror symmetry
of the geometry of Calabi-Yau moduli spaces. In this paper we shall consider the implica-
tions of mirror symmetry of the spectrum of BPS soliton states, which are associated to
minimal cycles in the Calabi-Yau. New surprises will be found.
The basic idea we will investigate is briefly as follows. Consider IIA string theory
compactified on a large Calabi-Yau space X. In four dimensions there are BPS states
arising from the reduction of the ten-dimensional 0-brane. The moduli space of this 0-
brane is X itself. In four dimensions the 0-brane can be described by a supersymmetric
worldbrane sigma model with target X. The BPS states are the cohomology classes on X,
and interactions will involve other invariants associated to X.
Quantum mirror symmetry1 implies that an identical theory is obtained by compacti-
fying the IIB theory on the mirror Y of X. In this formulation of the theory, all BPS states
arise from supersymmetric 3-branes wrapping 3-cycles in Y . (The condition for supersym-
metry is [4] that the 3-cycle is a special lagrangian submanifold [6] and the U (1) connection
on the 3-cycle is flat.) Hence there must be such a 3-brane in Y whose moduli space is
X.2 The 3-brane moduli space arises both from deformations of the 3-cycle within Y as
well as the flat U (1) connection on the 3-cycle.3 Both of these are generated by harmonic
1-forms on the 3-cycle [7], and the real dimension of the moduli space is accordingly 2b1 .
Since this moduli space is X this 3-brane must have b1 = 3 in order for the dimensions
to match.4 Now the moduli space of the flat connections on a 3-cycle with b1 = 3 at a

1
This goes beyond the usual assertion that the conformal field theories are equivalent, and
requires that the full quantum string theories are equivalent [2][3][4][5].
2
Equality of the BPS spectrum only implies that the cohomologies are the same, but equality
of the full theory, as well as their perturbation expansions, with all interactions will require that
the spaces are actually the same. For example if X is large we can localize the 0-brane near a
point in X at a small cost in energy. For very large X the space of such configurations approaches
X. Such strong relations are not possible e.g. in the context of heterotic-IIA duality because the
perturbation expansions are not directly related.
3
As discussed in section two, a careful definition of the geometry involves open string
instantons.
4
It should also have no self-intersections because massless hypermultiplets at the intersection
point have no apparent analog on the 0-brane side.

1
fixed location in in Y is a three-torus which, as we shall see in the following, is itself a
supersymmetric 3-cycle in X. Hence this construction describes X as a three-parameter
family of supersymmetric three-tori. We will refer to this as a supersymmetric T 3 fibration
of X, which in general has singular fibers (whose nature is not yet well-understood, but
is constrained by supersymmetry). Consideration of IIB rather than IIA on X similarly
yields a description of Y as a family of supersymmetric three-tori. Hence the 3-cycles with
b1 = 3 must be three-tori.
To summarize so far, mirror symmetry of the BPS states implies that every Calabi-
Yau X which has a mirror has a supersymmetric T 3 fibration.5 The moduli space of the
supersymmetric 3-tori together with their flat connections is then the mirror space Y . Note
that this is an intrinsic formulation of the mirror space.
Now consider the action of T-duality on the supersymmetric T 3 fibers. A 0-brane
sitting on the fiber will turn into a 3-brane, while a 3-brane will turn into a 0-brane. T-
duality does not change the moduli space of the D-brane so this is the same 0-brane -
3-brane pair discussed above. We conclude that mirror symmetry is nothing but T-duality
on the T 3 fibers!
We will provide a check on this last statement in a certain limit in section two by an
explicit local computation of both the T-dual and moduli space geometries. The equality
of these two geometries will be seen to follow from the fact that supersymmetric 3-cycles
are volume-minimizing. We also discuss the interesting role of open-string disc instantons
in correcting the moduli space geometry.
The preceding arguments assume quantum mirror symmetry. It is important to un-
derstand how much of this structure can be directly derived without this assumption. As a
step in this direction in section three we show directly that the moduli space is Kahler and
that it is endowed with a holomorphic b1 form, in agreement with conclusions following
from mirror symmetry. Section four contains some examples. Concluding remarks, in-
cluding a prescription for constructing M-theory duals for a large class of N = 1 heterotic
string compactifications, are in section five.

5
This is reminiscent of the situation found for other types of dualities. For example it is
suspected that a type II compactification has a dual heterotic representation if and only if the
Calabi-Yau admits a K3 fibration [8][9].

2
2. Mirror Symmetry and T-Duality

Consider a family of toroidal, supersymmetric 3-branes L in a Calabi-Yau space X

endowed with metric gM N and 2-form BM N . These are defined by the map

X M (σ i , ta ) (2.1)

where σ i , i = 1, 2, 3 is a periodic spatial coordinate on L and ta , a = 1, 2, 3 is a coordinate

on the moduli space of supersymmetric maps. There is also a U (1) connection A on the
3-brane:
A = ai dσ i (2.2)

Supersymmetry requires [4] that (i) the connection A is flat (ii) the pullback of the Kahler
form on to the 3-cycle vanishes (iii) the pullback of the holomorphic 3-form on to the
3-cycle is a constant times the volume element. These last two conditions imply that the
3-cycle is a special Lagrangian submanifold [6](cf. section four also).
We are interested in the full moduli space M, parametrized by (a, t), of supersymmet-
ric 3-branes with flat connections. We will find it convenient to choose local coordinates

X i = σ i , X a+3 = ta , a = 1, 2, 3. (2.3)

The line element on X is then

ds2X = gab dta dtb + 2gai dta dσ i + gij dσ i dσ j . (2.4)

(2.4) locally describes a T 3 fibration of X. The two form is

1 1
B= Bab dta ∧ dtb + Bai dta ∧ dσ i + Bij dσ i ∧ dσ j . (2.5)
2 2
In general there will of course be singular points at which the fibers degenerate.
We wish to compute the metric on M. We first consider the tree-level contribution.
This is derived from the Born-Infeld action for the 3-brane [10,11]
Z p
S3 = − d4 σ −det(E − F ). (2.6)

where
Fµν = ∂µ Aν − ∂ν Aµ ,
Eµν = EM N ∂µ X M ∂ν X N ,
(2.7)
EM N = gM N + BM N ,
µ, ν, α, β = 0, 1, 2, 3,

3
and we have fixed the string coupling.
To describe slowly time-varying configurations we insert the ansatze

Ẋ a+3 = ṫa ,
Ẋ i = −E ji Eaj ṫa , (2.8)
F0i = ȧi ,

where the dot denotes differentiation with respect to time σ 0 , and E ij Ejk = δ i k .6 The Ẋ i
term is arranged so that the motion of the 3-brane is normal to itself, with a B-dependent
rotation in L. Expanding (2.6) to quadratic order in time derivatives, the leading kinetic
part of the action is then
Z
√
S3 = d4 σ −E[(Eab − Eai E ij Ejb )ṫa ṫb
(2.9)
+ (E ij Eja − Eaj E ji )ȧi ṫa + E ij ȧi ȧj )] + ....

Hence the tree-level line element on M is

ds2M = (Eab − Eai E ij Ejb )dta dtb + (E ij Eja − Eaj E ji )dai dta + E ij dai daj . (2.10)

Mirror symmetry implies that the full metric on M equals the metric on the mirror
Y of X. The leading result (2.10) can not be the full story in general. The metric (2.10)
has a U (1)3 isometry generated by shifts in ai . This is certainly not present for a generic
Calabi-Yau, so (2.10) can not in general be the exact metric on the mirror. However there
are instanton corrections which can resolve this apparent discrepancy. The motion of the
3-brane is generated by open strings with Dirichlet boundary conditions on the 3-brane.
The metric on M is defined by a two point function on a disc whose boundary is the
3-brane. This receives instanton corrections from minimal area discs whose boundaries
wrap a nontrivial 1-cycle in L. (If Y is simply connected all such cycles bound discs.)
Generically such corrections become exponentially small at large radius. However even at
large radius the corrections are non-negligible near the singularities of the fibration where
the T 3 degenerates. After including these corrections mirror symmetry predicts that (2.10)
will agree with the metric on Y . This is quite analogous to the usual mirror story where

6
Note that E ij is the inverse of the tensor Eij on L. So, for example, E ia makes no sense,
and E ij Eja 6= 0 in general.

4
the flat metric on the complexified Kahler cone agrees with the metric on the space of
complex structures only after including world sheet instanton corrections.
Instanton corrections to the moduli space geometry of 3-cycles in X are suppressed
in the limit of large radius away from the singular fibers. The moduli space geometry is
then the large complex structure limit of Y . In this limit the metric takes the symmetric
form (2.10) and hence admits a local U (1)3 action away from the singular fibers. We refer
to this as a “semi-flat” Calabi-Yau metric. A lower-dimensional example of such a metric
is found in equation 4.1 of [12], where stringy cosmic strings are described as a flat T 2
fibered over R2 with a singularity at the origin.7 A semi-flat K3 metric can be obtained
by patching together 24 such singular cosmic strings. A 6-dimensional example can be
obtained by simply tensoring this K3 with T 2 . The semi-flat metric on the quintic will be
discussed in section four.
Because there is locally a U (1)3 symmetry, an equivalent string theory can be explicitly
constructed locally (away from the singularities) by T-duality on the flat T 3 fibers. This
duality interchanges IIA and IIB, and maps IIA 0-branes and supersymmetric IIB 3-branes
tangent to the fibers into one another. Hence this T-duality is the same as mirror symmetry.
It is illuminating to check in detail by a local calculation that the T-dual and moduli
space geometries are indeed the same. The string sigma model action on a flat worldsheet
is Z
1 ¯ b + Eia ∂σ i ∂t
¯ a + Eai ∂ta ∂σ
¯ i + Eij ∂σ i ∂σ
¯ j ).
Sσ = (Eab ∂ta ∂t (2.11)
2π
The three symmetries of shifts of the σ i imply that the three currents

ji = Eji ∂σ j + Eai ∂ta (2.12)

are conserved. We may therefore define three real worldsheet scalars ai by

ji = i∂ai . (2.13)

Eliminating the scalars σ i in favor of ai one finds an equivalent T-dual sigma model.
The T-dual metric is precisely (2.10), see for example [13] where various subtleties in the
transformation are also discussed.
7
The resolution described in [12] of those singularities can be understood as arising from disc
instanton corrections.

5
 In general the T-dual of a Ricci-flat metric obeys the low energy string equations
of motion. However it is not necessarily itself Ricci-flat because T-duality generically
generates an axion field strength H = dB and a dilaton field Φ with a nonvanishing
gradient, both of which act as a source for the Ricci tensor. For simplicity consider in the
following the special case in which B is zero before T-dualizing. The new Φ is then

Φ = ln detgij . (2.14)

The new H is
H i ab = g ij ,a gjb + g ij gjb,a − g ij ,b gja − g ij gja,b . (2.15)

The metric g on X in the coordinates (2.4) obeys various differential identities by virtue
of the fact that all three-tori at constant ta minimize the volume [4]. These constraints on
g imply that H vanishes and Φ is constant after T-duality, as required.
To see this consider a small normal perturbation of the three-torus at t = 0 described
by
X i = σ i − g ij (0)gja (0)Λa (σ),
(2.16)
X a = Λa (σ).
Through first order in Λ and first order in ∂Λ the induced metric h is given by, in the
semi-flat case where the metric is independent of σ,

hij = gij + gij,a Λa + gia,b Λb ∂j Λa + gja,b Λb ∂i Λa
 (2.17)
− gil,b g lk gka Λb ∂j Λa − gjl,b g lk gka Λb ∂i Λa t=0 .

The change in the volume may be written, using (2.14) and (2.15) and integrating by parts
Z √
δV3 = d3 σδ h
Z (2.18)
1 √

= d3 σ g ∂a ΦΛa + H i a
ab Λ ∂i Λ .
b
2

The fact that the area does not change at first order in Λ implies that the dilaton is
constant. The Λ∂Λ variation will be non-negative for all Λ if and only if H = 0. Hence H
vanishes and the T-dual metric must be Ricci-flat.
At this point it is easy to see that, as mentioned in the introduction, the three-tori
defined by t = constant in the moduli space metric (2.10) are themselves supersymmetric
3-cycles in the mirror Y of X. One way of stating the condition for supersymmetry is that

6
the T-dual geometry should have vanishing H and constant Φ. This is obviously the case
since T-dualizing just returns us back to X.
Now consider perturbing the Calabi-Yau moduli so that the metric is no longer semi-
flat. In that case the action of T-duality on the fibers still exists but is complicated be-
cause there are no isometries. The construction of the moduli space M of supersymmetric
3-branes is complicated by instantons. Nevertheless since they are both described as per-
turbations of the same theory the equivalence between T 3 T-duality and mirror symmetry
should be valid for all Calabi-Yau geometries in a neighborhood of the semi-flat geome-
tries. Furthermore since the supersymmetric 3-cycles correspond to minimally-charged
BPS states we expect that they survive sufficiently small perturbations. In conclusion
the relation between fiberwise T-duality and mirror symmetry is quite general and not
restricted to the semi-flat limit.

3. D-brane moduli space

We now turn to some mathematical generalities of the above discussion. We will

not be able to prove all the statements of the previous section regarding D-brane moduli
space, but we hope to provide a framework for future work on these matters. We will
show that with a certain natural metric, the D-brane moduli space has a Kähler structure.
In addition, there are several natural forms on this moduli space which suggest that its
interpretation as a Calabi-Yau for toroidal D-branes makes good sense. In fact, a “dual”
geometry emerges which corresponds to the mirror symmetry interpretation of the previous
section.
Recall that we are interested in a special Lagrangian submanifold, L, of a Calabi-Yau
manifold, M. Let f : L → M be the map of the imbedding (generally, f will only need to
be an immersion). The special Lagrangian condition is equivalent ([14], Corollary 7.39) to
the statement that
f ∗ω = 0 and f ∗ κ = 0,

where ω is the Kähler form and κ is the imaginary part of the Calabi-Yau form on M. The
asterisk denotes the pull-back operation. We will study the space of special Lagrangian
immersions f (which arise as deformations of a fixed immersion) and call this space Msl .
We use the following conventions:
• α, β, γ, ... are indices for M.

7
 • y α , y β , ... are coordinates on M.8
• i, j, k, ... are indices for L.
• xi , xj , ... are coordinates on L.
• Msl is the moduli space of special Lagrangian submanifolds.
• a, b, c, ... are indices on Msl
• ta , tb , ... are coordinates on Msl .
• gαβ , gij , gab are metrics on M, L, Msl .9 Without indices, we write g, g, gSL.
• J α β is the complex structure on M. J α γ J γ β = −δ α β .
• ωαβ = gαγ J γ β .
∂f α
• f αi = ∂xi , etc.
• All indices are raised or lowered by the appropriate metric.
∂
• For a Lagrangian submanifold, a tangent vector wi ei (where ei = ∂xi
) determines
a normal vector w = wi (J · f∗ ei ), and its associated second fundamental form (hw )jk is
given by wi hijk , with h a symmetric tensor defined by hijk = −h∇f∗ ei f∗ ej , J · f∗ ek i [15].
As yet, we have not defined the coordinates on moduli space, or even shown that it
exists as a manifold.10 We do so now, defining coordinates analogous to Riemann normal
coordinates, which give a coordinatization of a Riemannian space by the tangent space of
a point. Our point will be the initial special Lagrangian submanifold, L, with the map f.
The tangent space to Msl at L can be found by considering an arbitrary one-parameter
deformation f (t) of f = f (0). In order to remove ambiguity, we require throughout that
f˙, a vector field on f (L) in M, be a normal vector field. This can be achieved through a
df α ∂
t-dependent diffeomorphism of L. Now f˙ = dt ∂y α
is orthogonal to f (L). Therefore, J · f˙
is a tangent vector (by the Lagrangian condition), which we can convert to a 1-form by
df β α
the metric g on M. This combination gives us a 1-form dt ωβα dy . We define θ to be the
pull-back of this 1-form to L; θi = f˙β ωβα f α i .11

8
We will not use complex coordinates for M ; we will use real coordinates adapted to the
complex structure, which will then be constant in these coordinates.
9
This notational redundancy begs the question, “What is g12 ?” We will be careful never to use
numbers where our indices stand, and operate under the assumption that an additional identifier
would be employed (e.g., g 12 on M ).
10
This was shown on general grounds in [7], though without the introduction of coordinates,
which we shall need for our geometric analysis.
11
Note that the definition of θ is independent of the requirement that f˙ be normal, for any
tangent vector added to f˙ adds nothing to θ, by the Lagrangian condition.

8
 d ∗
Proposition: dt f ω = dθ.
Proof: Commuting derivatives and using the closure of ω, one computes

d d

f (t)∗ ωij (x) = f α i f β j ωαβ (f (x))
dt dt
= ∂i (f )f j ωαβ + f α i ∂j (f˙β )ωαβ + f α i f β j f˙γ ∂γ ωαβ
˙ α β

= ∂i (f˙α )f β j ωαβ + f α i ∂j (f˙β )ωαβ + f α i f β j f˙γ (−∂α ωβγ − ∂β ωγα )

 
˙ α β α ˙β β ˙γ
= ∂i (f )f j ωαβ + f i ∂j (f )ωαβ + f j f ∂i ωγβ − f i f ∂j ωγα α ˙γ

   
= ∂i f˙α ωαβ f β j − ∂j f˙β ωβα f α j
= (dθ)ij .

Therefore, to preserve the Lagrangian condition (to first order), the normal vector from
the deformation f (t) must give rise to a 1-form which pulls back to a closed 1-form θ on
L.
Likewise, we have the following proposition.
d ∗
Proposition: dt f κ = 0 ⇒ d† θ = 0.
The proof, similar to (though not quite as straightforward as) the above, may be found
in [7]. Therefore, θ must be closed and co-closed, i.e. harmonic.
The tangent space to deformations of isomorphism classes of flat U (1) bundles is
simply described. If A represents a flat U (1) connection on a bundle E → L, then a
deformation of the connection has the form A + θ. Requiring this to be flat means that
d(A + θ) = 0 and hence dθ = 0, since A is flat. Further, A (equivalently θ) is only defined
up to gauge transformations A ∼ A + dχ. Therefore, the tangent space to the space of flat
bundles is closed 1-forms modulo exact 1-forms, i.e. H 1 (L). A Riemannian metric on L –
in our case the pull-back metric – provides an isomorphism H(1) ∼ = H1 (L). In fact the full
space of flat U (1) bundles is described by the set of monodromies, exp[2πicj ], j = 1...b1 (L),
where b1 is the first Betti number. This space is the torus T b1 . The moduli space M is a
fibration over the moduli space of special Lagrangian manifolds (continuously connected
to f (0)(L)), with fiber T b1 .
Therefore, the tangent space to M at L is H1 (L) ⊕ H1 (L), and there is a natural
metric on H1 , which defines for us a metric on Msl and a block diagonal metric on M.
Let θ a and θ b be two harmonic 1-forms on L, with its induced metric gij . Then
 
gSL 0
gM = ,
0 gSL

9
where Z
√
(gSL )ab = θia θjb g ij gdx.
L

There is also a natural almost complex structure, which takes the form
 
0 1
J = .
−1 0

The coefficients of this matrix might not be constant in a coordinate frame for M, so we
have not yet shown that it is a complex structure, or that the metric is Kähler with respect
to it. We shall do this presently, after defining a coordinate frame.
Knowing that the tangent space to M at L is H1 ⊕ H1 is not enough for us to be able
to set up coordinates for M. To do this, we need something analogous to the exponential
map for Riemann normal coordinates. In this case, one needs, given a tangent direction,
a uniquely defined one-parameter family of flows of L in M which are special Lagrangian,
i.e. for which f (t)∗ ω = f (t)∗ κ = 0 for all t. Consider a harmonic 1-form θ on L, with w
its corresponding vector, wi = g ij θj . w obeys ∇i wi = 0. By continually pushing forward
w and rotating it by the complex structure, we can attempt to define a flow through the
equation
f˙(t) = J · f (t)∗ w.

However, there is no guarantee that the manifolds f (t)(L) will continue to be special
Lagrangian manifolds at all times. The problem is that g = f ∗ g evolves in t, and so θ
may not remain harmonic. To remedy this, we require θ to be harmonic by including a
time-dependence explicitly. We do so in such a way that [θ], the cohomology class defined
by θ, does not have a t-dependence. Therefore, we define

e = θ − dφ(t);
θ(t)

then the requirement that θe be harmonic becomes ∇i (g ij ∂j φ) = ∇i wi or

△φ = ∇i wi = d† θ.

Do not forget the implicit t-dependence of the operators! Note that at a given time t, the
last equation determines φ up to an additive constant, which is irrelevant to the definition
e
of θ(t).

10
 We can now try to define a one-parameter flow starting from L with initial normal
vector J · f (0)∗ w as the solution to the coupled equations

df α ∂f β
= J α β i g ij (θ j − ∂j φ)
dt ∂x (3.1)
†
△φ = d θ,

where θ has no t-dependence. This is analogous to geodesic flow on Riemannian manifolds

(though we don’t claim that this flow is geodesic with respect to the metric on moduli
space).
Proposition: This flow exists, at least infinitesimally, and is unique.
Proof: A proper demonstration of the above uses results we will derive in this section.
Basically, one can recursively solve for every derivative of f with respect to t. At first order,
1 †
the bottom equation specifies φ = △d φ + constant, which gives a unique expression for
d2 f
the top equation (the constant falls out). To find dt2
one can differentiate the bottom
equation to get △φ̇ = d† ψ, for some ψ (specifically, ψj = 2(hw )jk g kl (∂l φ − θl )), and so
on for the higher derivatives. We have not determined the radius of convergence (possibly
zero) for this expansion in t. For the remainder of this paper, we assume that this flow
exists for finite time.
Let θ a , a = 1...b1 be a basis for H∞ (L), wa the corresponding vectors, and ta the
corresponding coordinates in moduli space. We call Lt the submanifold f (1)(L), where
the flow is defined by the 1-form θ = ta θ a . We can coordinatize the full moduli space M
with ta and sa , a = 1...b1 , where the sa simply tell us to use the connection A + sa θ a on
Lt . Now while we know that θ does not change in cohomology during the flow it defines,
other 1-forms might do so. As a result, the almost complex structure, J , could pick up a
coordinate dependence.12 The calculation in the appendix will show that, to first order,
this does not occur, and as a result the Nijenhuis tensor vanishes. This shows that J is
indeed a complex structure on moduli space.
We turn now to the geometry of the flow. Let f (0) be an imbedding and f (t) a family
of special Lagrangian immersions inducing Then we have the following result.
d
Proposition: g
dta ij
= 2hijk wak , which is twice (hw )ij , the second fundamental form
defined by the normal vector J · w.
 
12 b b b c 0 y−1
If [θ (t)] is not constant, write [θ (t)] = y c (t)θ . Then J = .
−y 0

11
Here
hijk = −h∇f∗ ei f∗ ej , J · f∗ ek i,

and ∇ represents the covariant derivative on M. It is not hard to show that by the La-
grangian property h is a symmetric three-tensor on L. These and other properties of
Lagrangian flow are investigated in [15]. The proof of the proposition is straightforward –
one simply puts gij = f α i f β j gαβ and differentiates, substituting (3.1) for time derivatives
(note that d gαβ = f˙γ ∂γ gαβ ).
dta
As with Riemann normal coordinates, we have to express the vector field defined
by θ b as a function of the coordinate t (analogous to push-forward by the exponential
map). As a result, the forms pick up a t-dependence. We hope to find that our flow
gives the θ a a t-dependence such that the cohomology class does not change as a function
of time, but note that the form itself must change, since it must remain harmonic with
d b
respect to the changing metric. That is, we want that dta [θ ] = 0, where the brackets
represent cohomology (the flows are already defined so that dtda [θ a ] = 0, no sum) – i.e.,
d b
that dta
θ = dψ ab for some function ψ ab (t; x). We cannot solve the flows for finite time, but
d b
we can compute dta θ |t=0 as follows. Consider a one-parameter family of flows fr (t) defined
a b
with initial vector θ + rθ . These define a two-dimensional cone in moduli space, and we
1 d 13
can take the derivative in the b direction at time ta by computing ta dr |r=0 . This gives
a normal vector (it may not be normal, but the pull-back to a 1-form will be insensitive
to tangent directions), which then defines a 1-form representing θ b (ta ). Specifically,
 
1 dfrβ ∂f γ
b
θ (ta )i = − J α β |r=0 gαγ i , (3.2)
ta dr ∂x

where the family of flows is defined by the equations

df α ∂f β
= J α β i g ij (θja + rθjb − ∂j φ)
dt ∂x
△φ = d (θ + rθ b ).
† a

Do not forget the r and t dependence hidden in the second equation. Note, too, that θ a
and θ b have no t- or r-dependence.
We can see that limta →0 θ b (ta ) = θ b (0), as it must, by L’Hôpital’s rule and some
algebra, noting that f (0) is independent of r.

13
If we write ta , we mean to set ~t = (0, ..., ta , ...0), with lone nonvanishing ath component.

12
 b
It remains to compute limta →0 dθdt(t)
a
. This computation, performed in the appendix,
depends on using L’Hôpital’s rule and applying the definition of the flow for each time
derivative. After some algebra, one finds
 
dθ b (ta ) 1 d2 φ
|t=0 = − d ≡ dψ ab . (3.3)
dta 2 drdt r=t=0

This shows, too, that

Z
d √
∂a gbc = θ b (t)i θ c (t)j g ij gdx
dt
Z a Z Z
√ (3.4)
= hdψ , θ i + hθ , dψ i − 2 hijk wai wbj wck gdx
ab c b ac

= ∂b gac ,

where we have used harmonicity and the symmetry of hijk , as well as the fact that the
√
derivative of g is proportional to the mean curvature, which vanishes for minimal mani-
folds (h is traceless). The result (3.3) also shows that the derivatives of J are zero, which
immediately implies that it is a complex structure.14 In addition, (3.4) tells us that the
M
Kähler form on moduli space, ωab = gac J c b , is closed at L, which was an arbitrary point
on M.
On Msl , there exists a natural n-form (where 2n is the real dimension of M,) Θ,
defined as follows [7]: Z
a1 an
Θ(θ , ..., θ ) = θ a1 ∧ ... ∧ θ an .
L

Now since in our coordinates the θ ai change by exact terms, all derivatives of Θ vanish,
and Θ is closed. Further, we can extend Θ to a form on M by defining,

ΘM = Θa1 ...an dz a1 ...dz an ,

where dz a = dta + idsa . The assertion that dΘM = 0 follows from the above, and fact that
there is no s-dependence. We could have complex conjugated some of the dz’s and obtained
other forms. We don’t know how many of these forms will survive the compactification
procedure, but we point out that in the special case of toroidal submanifolds, b1 = n and
so ΘM is a holomorphic b1 -form, which could be the Calabi-Yau form. This lends support,
at least, to the conjectures of the previous section. Note, too, that one can also, given a
14
The Nijenhuis tensor, N c ab = J d a (∂c J c b − ∂b J c d ) − (a ↔ b), clearly vanishes.

13
 R √
2-form potential Bαβ , define a 2-form on moduli space Bab = B ij θia θjb g, where Bij are
the components of f ∗ B, and coordinates are raised, as usual, by g ij .
Of course, there is more about this moduli space to study. We need to determine
its curvature. The variation of the metric means that our coordinates are not normal. A
coordinate transformation mixing t and s is needed to make it so. Therefore, the fibration
may not be holomorphic with respect to this metric. This makes some sense, as we don’t
expect, for example for K3, to find the fibers always perpendicular to the base. However, we
do not yet know how this metric should get corrected, or what the proper compactification
of moduli space looks like. We do not know how the Kähler class gets corrected, as a
result. We hope more rigorous mathematical results will ensue.

4. Examples

4.1. K3

The only well-understood example of the preceding is K3 (or K3 × T 2 , if we want

a three-fold). The supersymmetric cycles can be related, by a rotation of complex struc-
tures15 to Riemann surfaces sitting in the K3. For genus one, these are tori, which are
easily apparent if the K3 is written as an elliptic fibration: π : K3 → P1 . Each point p on
the K3 determines a torus π −1 (π(p)), and exists as a point on that torus (which is its own
Jacobian). Given a section, this uniquely determines a submanifold and a flat bundle on it.
Thus K3 appears as the moduli space for submanifolds which are tori. This is appropriate,
as K3 is its own mirror!

4.2. The Quintic

The example of the quintic has been analyzed in great detail [16]. It can be seen as
follows that, at least for sufficiently large complex structure, there is a supersymmetric
T 3 cycle in the homology class predicted by mirror symmetry. On the mirror quintic Q
there is a a 0-brane, with moduli space equal to Q. There is also a 6-brane which wraps
Q and has no moduli (note that there are no flat bundles since Q is simply connected).

15
The K3 has three complex structures, I, J, K, giving three 2-forms, ωI , ωJ , ωK which can be
interchanged by SO(3) rotations. With complex structure J, we have ω = ωJ and ΩJ = ωI + iωK ,
so we can relate the condition f ∗ (ω) = f ∗ (ImΩJ ) = 0 with complex structure J to f ∗ (ΩK ) = 0,
in complex structure K. These surfaces are then holomorphic submanifolds with respect to K.

14
Quantum mirror symmetry tells us that the corresponding 6- and 0-cycles should be mirror
e The mirror transformations relating the even
to two distinct 3-cycles on the quintic, Q.
e have been explicitly displayed in [16]. Using
cohomology on Q to the odd cohomology on Q
these transformations it was shown in [17] that the 6-cycle on Q is mirror to the 3-cycle
e which degenerates at the conifold. Since the 0-cycle is Poincare dual to the 6-cycle
on Q
(at large radius where instanton corrections can be ignored) it must be mirror to a 3-cycle
Poincare dual to the one which degenerates at the conifold. Consider the 3-cycle defined
in [16] by
|z1 | = a, |z2 | = b, |z3 | = c, (4.1)

where a, b, c are real. Consider a patch in which we can fix z5 = −1 and take z4 to be a
root of the polynomial equation for a quintic

z15 + z25 + z35 + z45 = 1. (4.2)

If a, b, c are sufficiently small the roots are non-degenerate and as the phases of z1 , z2 , z3
e Topologically (4.1) defines a T 3 “fibration” of Q
vary a three-torus is swept out in Q. e with

base parameterized by a, b, c. The fibers become singular at values of a, b, c for which

the roots collide. It was shown in [16] that this 3-cycle is Poincare dual to the vanishing
cycle at the conifold. Hence it is in the correct homology class to be mirror to the 0-brane
on Q. To see that there is actually a supersymmetric cycle in this class consider the large
complex structure limit in which a term λz1 z2 z3 z4 z5 is added and the coefficient λ taken
to infinity. There is a branch on which we may take z5 = −1 and z4 near zero and the
metric approaches (up to a constant)

ds2 = d ln z1 d ln z̄1 + d ln z2 d ln z̄2 + d ln z3 d ln z̄3 (4.3)

for finite z1 , z2 , z3 . In this limit (4.1) obviously defines a family of supersymmetric 3-

cycles. Perturbing away from the large complex structure limit produces a small effect
on the Ricci-flat metric for finite z1 , z2 , z3 . Since supersymmetric 3-cycles are expected
to be stable under small perturbations of the metric, such a 3-cycle should exist in a
neighborhood of the large complex structure limit.

15
5. Conclusions

Our results should lead to the construction of a large new class of dual pairs. For
example, heterotic string theory on T 3 is equivalent to M-theory on K3. Applying this
duality fiberwise to the T 3 fibers of any N = 1 compactification of heterotic string theory
on a Calabi-Yau space with a mirror partner, one obtains an N = 1 compactification of
M-theory on a seven-manifold of G2 holonomy. Examples of this type are in [18][19]. By
considering perturbations this duality can be extended to (0, 2) heterotic compactifica-
tions.16 Perhaps this will provide a useful way to study a phenomenologically interesting
class of string compactifications. A precise understanding of this duality, as well as the
mirror relation discussed herein, will probably require a better understanding of the nature
of the singular fibers. Undoubtedly, constraints from supersymmetry will play a role in
controlling these singularities.

In this paper we have only considered the implications of quantum mirror symmetry
for the simplest case of a single 0-brane. We expect that consideration of other p-branes
and their bound states will lead to further insights into the rich structure of Calabi-Yau
spaces and supersymmetric string compactifications in general.

d b
Appendix: The calculation of dta θ (0).

d b
We calculate dta
θ (0), where conventions are as in section three. Let us now use t for
ta . We have from (3.2) that θ b (t) is of the form θ b (t) = C/t, where C → 0 as t → 0 (since
dfs
ds = 0 at t = 0). So

dθ b
limt→0 = limt→0 (tĊ − C)/t2
dt
= limt→0 (tC̈ + Ċ − Ċ)/2t by L’Hôpital’s rule
= limt→0 C̈/2 = C̈(0)/2.

16
Recent progress on this problem has been made using F-theory duals in [20].

16
 β
Let’s calculate. θ b (t)i = − 1t J α β dfds gαγ f γ i .
 2 β
b αd γ d df β 1 α γ d d f
θ̇ (0)i = −J β (gαγ f i ) − gαγ J β f i
dt ds dt 2 ds dt2
d  β µ kl a 
= −J α β (∂ρ gαγ )(J ρ τ f τ j waj )f γ i J µ f k g (θl + sθlb − ∂l φ)
ds
d  
− J α β gαγ ∂i (J γ µ f µ j waj ) J β ν f ν k g kl (θla + sθlb − ∂l φ)
ds
 2 β
1 d d f
− gαγ J α β f γ i
2 ds dt2
= −J α β (∂ρ gαγ )J ρ τ f τ j waj f γ i J β ν f ν k g kl θlb
− J α β gαγ J γ µ (f µ ij waj + f µ j ∂i waj )J β ν f ν k g kl θlb
 2 β
1 α γ d d f
− gαγ J β f i .
2 ds dt2

In the above and what follows we use the fact that derivatives of indices on M with respect
df α
to t can be obtained by dt α
∂ . Also, we will freely use that J α γ J γ β = −δ α β and that gαb
is hermitian: J α µ gαβ J β ν = gµν . Derivatives are finally evaluated at t = s = 0.

θ̇ b (0)i = (∂ρ gαγ )J ρ τ f τ j waj f γ i f α k wbk − gβµ J β ν (f µ ij waj + f µ j ∂i waj )f ν k wbk

 2 β
1 α γ d d f
− gαγ J β f i
2 ds dt2
= (∂ρ gαγ )J ρ τ f α k f τ j f γ i wbk waj
+ ωνµ f ν k f µ i jwbk waj
+ ωνµ f ν k f µ j (∂i waj )wbk
 2 β
1 α γ d d f
− gαγ J β f i
2 ds dt2

The third line in the last equality is zero since ωνµ f ν k f µ j equals f ∗ ωkj , which vanishes.

θ̇ b (0)i = J ρ τ (∂ρ gαγ )f α k f τ j f γ i wbk waj + ωνµ f ν k f µ j (∂i waj )wbk

 2 β
1 γ d d f
+ ωβγ f i .
2 ds dt2
 2 β
d d f
At this point we’ll need to calculate ds dt2 |t=s=0 . We have

df γ
= J γ τ f τ l g lm (θm
a b
+ sθm − ∂m φ).
dt
17
Differentiating again, we get
d2 f γ ∂


2
= J γ τ l J τ ν f ν n g np (θpa + sθpb − ∂p φ) g lm θm
a b
+ θm − ∂m φ
dt ∂x

a

+ J τ f τ l −2ha lm − 2shb lm θm
γ
+ sθm b
− ∂m φ
dφ
+ J γ τ f τ l g lm (−∂m ).
dt
Taking the derivative with respect to s gives (note φ = 0 at t = 0, independent of s):
 
d d2 f γ


2
= J γ τ ∂l J τ ν f ν n g np θpa g lm θm
b
+ J γ τ ∂l J τ ν f ν n g np θpb g lm θm
a
ds dt t=s=0
− 2J γ τ f τ l ha lm θm
b
− 2J γ τ f τ l hb lm θm
a
 2 
γ τ lm d φ
− J τ f l g ∂m
dsdt
= −∂l (f γ n wan )wbl − ∂l (f γ n wbn )wal
− 2J γ τ f τ l ha lm θm
b
− 2J γ τ f τ l hb lm θm
a
 2 
γ τ lm d φ
− J τ f l g ∂m .
dsdt

Let’s now plug this result into our formula for θ̇ b (0)i .

θ̇ b (0)i = J ρ τ (∂ρ gαγ )f α k f τ j f γ i wbk waj + ωνµ f ν k f µ j (∂i waj )wbk

1  
+ ωβγ f γ i −∂l (f γ n wan )wbl − ∂l (f γ n wbn )wal
2
1  
+ ωβγ f γ i −2J γ τ f τ l ha lm θm b
− 2J γ τ f τ l hb lm θma
2   2 
1 γ γ τ lm d φ
+ ωβγ f i −J τ f l g ∂m
2 dsdt
= J ρ τ (∂ρ gαγ )f α k f τ j f γ i wbk waj + ωνµ f ν k f µ j (∂i waj )wbk
1

− ωβγ f γ i f β ln wan wbl + f β ln wbn wal
2


− ωβγ J β τ f γ i f τ l ha lm θm b
+ hb lm θm
a
 2 
1 γ β τ lm d φ
− ωβγ f i J τ f l g ∂m
2 dsdt
= J ρ τ (∂ρ gαγ )f α k f τ j f γ i wbk waj + ωνµ f ν k f µ j (∂i waj )wbk
− ωβγ f γ i f β ln wal wbn


− gli ha lm θm
b
+ hb lm θm
a
 2 
1 d φ
− ∂i .
2 dsdt

18
 Note that gli ha lm θm
b
= gli hb lm θm
a
= hijk waj wbk . The rest is just algebra in combining
the other terms. It is trivial to show the following quoted result in normal coordinates on
the target space (which, since it is Kähler, can be chosen simul- taneously with coordinates
adapted to the complex structure), in which case hijk = ωαβ f α i f β jk . In any case, it can
be shown straightforwardly, in general. We get, since the h terms cancel, the following
remarkable result.  
dθib 1 ∂ d2 φ
|t=0 = − , (A.1)
dta 2 ∂xi dsdt
or
∂a θ b = dψ (A.2)
2
d φ
at t = 0, where ψ = − 12 dsdt |t=s=0 .

Acknowledgements
We would like to thank M. Bershadsky, D. Morrison, A. Sen, K. Smoczyk, P. Townsend
and C. Vafa for useful discussions. The research of A.S. is supported in part by DOE grant
DOE-91ER40618; that of S.-T.Y. and E.Z. by grant DE-F602-88ER-25065. A.S. wishes to
thank the physics and mathematics departments at Harvard, the mathematics department
at MIT, and the physics department at Rutgers for hospitality and support during the
course of this work.

19
 References

[1] A collection of introductory articles can be found in “Essays on Mirror Manifolds,”

S.-T. Yau, ed., International Press, 1992.
[2] A. Strominger, “Massless Black Holes and Conifolds in String Theory,” Nucl. Phys.
B451 (1995) 96, hep-th/9504090.
[3] P. Aspinwall and D. Morrison, “U-Duality and Integral Structures,” Phys. Lett. B355
(1995) 141, hep-th/9505025.
[4] K. Becker, M. Becker, and A. Strominger, “Fivebranes, Membranes, and Non-
Perturbative String Theory,” Nucl. Phys. B456 (1995) 130, hep-th/9507158.
[5] D. Morrison, “Mirror Symmetry and the Type II String,” hep-th/9512016.
[6] F. R. Harvey and H. B. Lawson, “Calibrated Geometries,” Acta Math. 148 (1982)
47.
[7] R. McLean, “Deformations of Calibrated Submanifolds,” Duke preprint 96-01:
www.math.duke.edu/preprints/1996.html.
[8] A. Klemm, W. Lerche, and P. Mayr, “K3 Fibrations and Heterotic-Type II String
Duality,” Phys. Lett. B357 (1995) 313, hep-th/9506112.
[9] P. Aspinwall and J. Louis, “On the Ubiquity of K3 Fibrations in String Duality,”
Phys. Lett. B369 (1996) 233, hep-th/9510234.
[10] R. G. Leigh, “Dirac-Born-Infeld Action from Dirichlet Sigma Model,” Mod. Phys.
Lett. A4 (1989) 2767.
[11] P. Townsend, “D-branes from M-branes,” hep-th/9512062.
[12] B. Greene, A. Shapere, C. Vafa and S.-T. Yau, “Stringy Cosmic Strings and Noncom-
pact Calabi-Yau Manifolds,” Nucl. Phys. B337 (1990) 1.
[13] A. Giveon and M. Rocek, “Introduction to Duality,” hep-th/9406178, to appear in
“Essays on Mirror Manifolds II,” International Press.
[14] F. R. Harvey, Spinors and Calibrations, Academic Press, New York, 1990.
[15] K. Smoczyk, “A Canonical Way to Deform a Lagrangian Submanifold,” dg-ga/9605005.
[16] P. Candelas, X. C. De La Ossa, P. Green, and L Parkes, “A Pair of Calabi-Yau
Manifolds as an Exactly Soluble Superconformal Theory,” Nucl. Phys. B359 (1991)
21.
[17] J. Polchinski and A. Strominger, “New Vacua for Type II String Theory,” hep-
th/9510227.
[18] J. Harvey, D. Lowe and A . Strominger, “N=1 String Duality,” Phys. Lett. B362
(1995) 65, hep-th/9507168.
[19] B. Acharya, “N=1 Heterotic/M-Theory Duality and Joyce Manifolds,” hep-th/9603033.
[20] E. Witten, “Phase Transitions in M-Theory and F-Theory,” hep-th/9603150.

20