SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING

SPACES

DANIEL M. HALMRAST

Abstract. Witten’s topological B-model on a Calabi-Yau background is known to reproduce,

in the open string sector, the derived category of coherent sheaves. When the target space is a
complex torus, the topological model enjoys a non-geometric symmetry known as T-duality, which
relates the theories on the torus and dual torus backgrounds. By considering the “double field
arXiv:2507.08181v2 [math.DG] 17 Jul 2025

theory” of Hull and Reid-Edwards on the product of a torus and its dual, T-duality occurs as a
geometric symmetry of the target space.
Building on the methods of Y. Qin, we propose a method of analyzing the topological B-model
on a torus in the doubled geometry framework which naturally incorporates certain rank-one D-
branes, providing a different perspective on the derived category of the torus. In certain cases, the
intersection theory of these lifted branes correctly computes the BRST cohomology of the B-model,
and hence the derived Hom-spaces of the corresponding line bundles.

Contents
1. Introduction 1
2. Physics Background 3
2.1. The Bosonic String 3
2.2. T-Duality of Strings on Tori 7
2.3. Supersymmetry and the Superstring 13
3. Mathematics Background 17
3.1. Connections on Holomorphic Line Bundles 17
3.2. Holonomy and Differential Characters 20
3.3. Complex Tori, Cohomology, and Line Bundles 24
3.4. Generalized Complex Geometry 32
4. Main Construction 34
4.1. The Doubling Torus 34
4.2. Lifting Generalized Complex Structures 35
4.3. Lifting D-branes to the Doubling Space 38
4.4. Geometry of Doubled Branes 42
5. Conclusions and Outlook 45
References 46

1. Introduction
During the second superstring revolution of the 90’s, many intricate and surprising relationships
were found between the physics of string theory and the mathematics of complex geometry. Wit-
ten’s topological string theory models [36] furnished a deep connection between the low-energy
physics of certain carefully-engineered string theory and certain invariants of the complex manifold
on which it propagates.
1
2 DANIEL M. HALMRAST

The result was extended to include exotic objects called “D-branes”, which appear as boundary
states in the open string theory. Although these naturally appear as coherent sheaves on the
target manifold, Sharpe [32] found that these D-branes may appear as more general objects in the
bounded derived category of coherent sheaves. The equivalence of categories between the category
of topological D-branes and the bounded derived category of coherent sheaves was constructed by
Aspinwall and Lawrence [3], and the predicted connection between the low-energy states of the
physical theory and the ext-groups appearing in the derived category was later computed by Katz
and Sharpe [24].
The dictionary between the topological nonlinear sigma model of [36] (and its categorization in
the sense of [3, 32]) and the complex geometry of the target space is not one-to-one. Rather, the
string theory carries additional structure that the target space geometry does not, and interpreting
this additional structure in terms of the geometry leads to many deep insights into the derived
category, including the far-reaching homological mirror symmetry conjecture of Kontsevich [25]
and the intricate structures of Bridgeland stability conditions [7], Donaldson-Thomas invariants
[26], and Joyce structures [21].
One such structure on the string theory appears in nonlinear sigma models into geometries
which are torus fibrations. The principle of “T-duality” states that the string theory on a torus
fibration is identical, via field redefinitions, to the string theory on the dual fibration. Using T-
duality, it is possible to construct a string theory which locally behaves like a nonlinear sigma
model into a torus fibration, but which admits no global geometric description [15]. The doubled
formalism, appearing concretely in [33][34], and formalized in [18], attempts to geometrize T-
duality by considering an associated string theory on the product of a torus and its dual.
Recently, the topological A-model was examined using this framework [31]. There, Y. Qin
proposed a way to double not only the ambient geometry but also the A-type D-branes on the
torus. He found that the intersection theory of the doubled D-branes is closely related to the
Fukaya category of the original torus.
In this paper, we propose an extension of the ideas of [31] to the topological B-model, and
explore the interplay between doubled geometry and generalized complex geometry. Section 2
covers the physics background which motivates the main construction. We review the basics of
the worldsheet perspective on string theory, and the interplay between the worldsheet perspective
and target space geometry. T-duality is introduced, and an example of a string theory with a non-
geometric background is examined in detail. We also see how supersymmetry on the worldsheet
constrains the geometry of the target space.
Section 3 covers the prerequisite mathematics for the main construction. Of primary importance
is the theory of connections on line bundles, which we approach from the analytic viewpoint.
The key tool is the theory of Cheeger-Simons differential characters, which is a refinement of
cohomology that tracks holonomy of U (1) connections. The theory simplifies beautifully on the
torus, and we analyze the analytic description of line bundles on tori, and their cohomology. Here
we introduce the notion of E-flatness, a condition on the connection of a complex line bundle. The
space of E-flat connections naturally forms a torsor over the dual torus, and has a particularly
nice description in terms of differential characters. Finally, we examine the theory of generalized
complex geometry in the sense of [12], and the connection between generalized submanifolds and
U (1)-bundles.

Main Results. The main results lie in section 4. We construct the doubling torus following the
ideas of [31], and propose a new definition of a lift L of E-flat rank-one space-filling D-branes in
the topological B-model. This lift is naturally expressed in terms of differential characters, and we
show that our definition of a lift agrees with [31]. These lifts are naturally holomorphic Lagrangian
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 3

subtori under the canonical structures on the doubling space, and we explicitly compute their
intersection properties.
A concrete link is then established between the generalized complex geometry of the base torus
X and the (ordinary) complex geometry of the doubled torus X.
Theorem (Theorem 4.3). The tangent bundle of X is canonically isomorphic (under σ) to the
pullback of the sum of tangent and cotangent bundles of either X or X̂. That is,
TX ∼
= π ∗ (T X ⊕ T ∗ X) ∼
= π̂ ∗ (T X̂ ⊕ T ∗ X̂) (1)
as smooth vector bundles.
We also establish that our definition of a lift L(L) of a line bundle L lifts the generalized tangent
bundle defined by the line bundle to the tangent bundle of the Lagrangian lift.
Theorem (Theorem 4.27). The tangent bundle of L(L) is the subbundle of T X defined by
T L(L) = {(v, v̂) ∈ T X ⊕ T X̂ | ιv̂ σ = ιv E} (2)
. That is, under the isomorphism T X ∼ = π ∗ (T X ⊕ T ∗ X) the tangent bundle of L(L) is isomorphic
to the pullback of the generalized tangent bundle of (X, E).
The main connection to physics is found in section 4.4.2, where we make explicit connections to
the derived category of the topological B-model. In particular, we establish agreement between a
certain holomorphic refinement of Floer intersection theory on the doubling space and the derived
Hom on the base.
Theorem (Theorem 4.36). Let L1 and L2 be two holomorphic line bundles with c1 (L1 ) = c1 (L2 ).
Then,
HomB (L1 , L2 ) = HF∗J (L(L1 ), L(L2 )) (3)
where HF∗J is the J -holomorphic part of HF∗ , HomB is the B-model open string spectrum, and
J = JJ is the complex structure on X induced by the complex structure on X (example 4.10).
Acknowledgements. This work is largely based off of my doctoral thesis [13] which was completed
at the University of California, Santa Barbara. I would like to thank in particular my advisor,
Dave Morrison, for teaching me all about the relevant string theory and complex geometry for
this problem, and for the invaluable insight and support throughout.

2. Physics Background
Despite the main construction of this paper being purely mathematical in nature, the motivation
of the construction stems mainly from notions in mathematical physics, namely string theory. We
will, therefore, take a detour through the landscape of string theory and the primary motivating
constructions found therein. We focus on two independent constructions: T-duality, a symmetry
of the string theory that is distinctly non-geometrical, and supersymmetry, which provides the link
between the physics and derived categories.
We follow the excellent exposition of [20] for this section.

2.1. The Bosonic String. In the worldsheet perspective on string theory, the fundamental space-
time is taken to be the string worldsheet, a two-dimensional (one spatial and one temporal) smooth
manifold Σ whose boundary components are all diffeomorphic to S 1 . It is imagined that the
boundary components are “incoming” and “outgoing” closed string states, and the field theory
computes the physics of the interaction described by the worldsheet topology.
4 DANIEL M. HALMRAST

2.1.1. The Free Bosonic String. The first physical theory to consider is the theory of a free scalar
boson on the worldsheet, which we review now. Take Σ = S 1 × R coordinatized by (σ, τ ), with σ
periodic of period 2π. Also, fix a pseudo-Riemannian metric γab on Σ for which σ is spatial and
τ is temporal. Then, let X : Σ → R be a scalar function on Σ, which we take to be a dynamical
object. These objects are governed by the Polyakov action, a local functional of γ, X and their
derivatives, defined as1
−1 √
Z
−γγ ab ∂a X∂b X dA


S[γ, X] = ′
(4)
4πα Σ
where α′ is a positive real number known as the “coupling constant”. The action governs the
entire quantum field theory, and if the action is left unchanged by some group action on X or γ,
that group action is said to be a symmetry of the system.
In fact, this action enjoys many symmetries.
• The group Diff 0 (Σ) of small diffeomorphisms of Σ acts locally on X and γ. Given a vector
field ξ a ,
X 7→ X + ξ a ∂a X
(5)
γ ab 7→ γ ab + ξ c ∂c γ ab − ∂c ξ a γ cb − ∂c ξ b γ ac
• The group C ∞ (Σ) acts on the metric by Weyl scaling. Given a function ω : Σ → R,
X 7→ X
(6)
γab 7→ e2ω γab

• The translation group R of R acts on the field X by the natural action. Namely, for a ∈ R,
X 7→ X + a
(7)
γ ab 7→ γ ab

To find the equations of motion of X, and hence the classical physics of this system, the
variational equation δS = 0 is imposed. Direct computation reveals that the fields X and γ
satisfy
√
−γ∇2 X = 0 (8)
. Using the action of Weyl scaling and Diff 0 (Σ), the metric γ can always be brought into standard
form
 
−1 0 ϕ
γ= e (9)
0 1
where eϕ is a global scaling function known as the conformal parameter. In this choice of metric,
the equations of motion become
∂σ2 − ∂τ2 X = 0


(10)
. This is easily solved using characteristics, from which we see that X decomposes into left-moving
and right-moving modes
X(σ, τ ) = XL (τ + σ) + XR (τ − σ) (11)

1We take throughout this section the convention of Einstein summation, where repeated indices are to be summed
over.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 5

. Imposing periodicity X(σ + 2π, τ ) = X(σ, τ ) allows the solutions to be expanded in a Fourier
series. Setting σ + = τ + σ and σ − = τ − σ,
 ′  12 X
− 1 ′ − α −
XR (σ ) = X0 + α PR (σ ) + i an e−in(σ )
2 2 n̸=0
 ′  12 X (12)
1 α 1 +
XL (σ + ) = X0 + α′ PL (σ + ) + i ãn ein(σ )
2 2 n̸=0
n

where X0 , PL,R are constants, and an and ãn are the Fourier coefficients. To enforce the reality
conditions XR∗ = XR and XL∗ = XL , the Fourier coefficients must satisfy
a−n = (an )∗
(13)
ã−n = (ãn )∗
. The complete solution is then
X = XL + XR = X0 + α′ (PL + PR )τ + α′ (PL − PR )σ + oscillators (14)
and single-valuedness around σ 7→ σ + 2π enforces PL − PR = 0.

2.1.2. Many Free Bosons and Maps to Target Space. Now that the theory of a single free boson has
been analyzed, the generalization to many non-interacting bosons is straightforward. As before,
Σ = S 1 × R is coordinatized by (σ, τ ), and γab is a pseudo-Riemannian metric on Σ. Now, fix a
positive integer d and let µ = 0, . . . , d − 1 be an index running from 0 to d − 1, and take d scalar
maps
Xµ : Σ → R (15)
as the bosonic fields on Σ. By adding their actions together, we obtain a new theory in which the
fields have no interaction with each other:
!
−1
Z X√
S [γ, X µ ] = −γγ ab ∂a X µ ∂b X µ dA (16)
4πα′ Σ µ

. The action enjoys all the same symmetries as the single free boson, with now d copies of the
Euclidean group of R:
• The small diffeomorphism group Diff 0 (Σ) acts locally as before.
• The Weyl group C ∞ (Σ) acts on the metric as before.
• The product of the translation groups of each bosonic field Rd acts by translation. Given
aµ ∈ R d ,
X µ 7→ X µ + aµ (17)
still leaves the action invariant.
Although this action is the most obvious one to incorporate many bosonic fields, the fact that it
is not invariant under the full Euclidean group of Rd , but rather only the translation subgroup,
suggests an extension.
Let gµν = δµν be the identity matrix. We can insert this into the action without modifying it
!
−1
Z X√
S [γ, X µ ] = −γγ ab ∂a X µ ∂b X ν gµν dA (18)
4πα′ Σ µ
6 DANIEL M. HALMRAST

. This new action is exactly equivalent to the old action. Now, however, we allow g to transform
under E(d) in exactly the opposite way X µ does, resulting in a E(d)-invariant theory. Specifically,
for Λµν an SO(d)-transformation,
X µ 7→ Λµν X ν
(19)
gµν 7→ (Λ−1 )ρµ (Λ−1 )σν gρσ
. This new SO(d)-action leaves the string action (18) unchanged, so the symmetry group of this
theory is enhanced to include all of E(d) instead of the subgroup of translations.
We emphasize that this model was built from worldsheet considerations, and is only a theory
of free scalar fields on a two-dimensional surface. From the perspective of the worldsheet, the
physics is that of d independent, non-interacting, chargeless distinguishable massless fields. With
our careful insertion of the identity matrix g and extension of the action of SO(d), the theory now
has a more “geometric” interpretation.
Let us now make the leap in identifying X µ with coordinates of an immersion of Σ into Rd ,
thought of as a Riemannian manifold with the standard flat metric. Fixing the standard basis for
Rd , define
φ : Σ → Rd
(20)
φ(σ, τ ) = X 1 (σ, τ ), . . . , X d (σ, τ )

. With this definition, the action (18) becomes

−1 √
Z
ab µ ν


S[γ, φ] = −γγ ∂ a φ ∂ b φ g µν
4πα′ Σ
(21)
−1 √
Z
ab ∗


= −γγ (φ g) ab
4πα′ Σ
where φ∗ g is the pullback of the metric to the worldsheet. This action is no longer dependant
on the choice of orthonormal frame X µ , but rather only on the map φ. Explicitly, due to the
E(d)-invariance of the action, switching coordinate systems to a new orthonormal coordinate
system
X µ 7→ Λµν X ν + aµ (22)
for any Λ ∈ SO(d) does not change the action. Equivalently, we were able to write the action in
a coordinate-free manner in (21) precisely due to this coordinate invariance.
With the action in the form of (21), it is apparent that the metric need not be the homogeneous
flat metric on Rd . Any choice of metric g on Rd (thought of as a Riemannian manifold) results in
a well-defined action.

2.1.3. Descent to a Manifold. The previous construction reinterprets the theory of d free bosons
as a theory of maps from Σ into Rd . The construction did not depend on the global structure of
Rd , and was also local on Σ as well. Instead of treating the model as a global map into Rd , let us
instead take everything to be done locally, and examine what happens.
Let (X, g) be a Riemannian manifold, and
` ια
α Uα X (23)
an open covering of X by coordinate charts. Define Uαβ to be the intersection of Uα with Uβ in
Uα . For any map φ : Σ → X, define
φUα : φ−1 (Uα ) → Uα (24)
to be the restriction of φ to a coordinate chart.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 7

On each φ−1 (Uα ), φUα maps into an open subset of Rd , and we can attempt to impose the
familiar action of (21). To keep things local, we instead define the Lagrangian density on this
open set as follows. Choose orthonormal coordinate functions xµ on Uα and define φµ := xµ ◦ φµUα
to be the local µth coordinate of φ. Then, define the local Lagrangian density to be the local
worldsheet 2-form
√
Lα [φUα ] = −γγ ab ∂a φµ ∂b φν gµν dA (25)
. By nature of the E(d)-invariance of this functional, this two-form is independent of choice of
coordinates. In fact, by Diff 0 (Σ)-invariance this is also independent of choice of coordinates on Σ.
In particular, this means that on the intersections Uαβ on the target and φ−1 (Uαβ ) on Σ the
worldsheet two-forms Lα and Lβ agree. Using this, we can glue the local densities into a global
2-form L[φ] on all of Σ. Then, we can define the bosonic nonlinear sigma model action to be
−1 √
Z
S[γ, φ] = ′
−γγ ab (φ∗ g)ab dA (26)
4πα Σ
.
Let us reflect for a moment on this construction. The starting point was a free field theory
on the worldsheet Σ, where we thought of the Σ as space-time and the field as a particle on
Σ. However, by careful manipulation of the action, we found that a theory of d distinguishable
bosonic particles on Σ admits the Euclidean group E(d) of Rd as a symmetry. This symmetry
allows us to reinterpret the worldsheet theory as instead a theory governing immersions of Σ into
Rd via the action (21). Finally, using the symmetry of the Euclidean group and, in particular,
the target space coordinate-invariance of the action, the theory could be extended to describe
immersions of Σ into arbitrary Riemannian manifolds. From the worldsheet perspective, these
are simply theories of d distinguishable bosonic particles propagating on Σ. However, from the
target-space perspective of the Riemannian manifold X, this is the theory of a propagating string
immersed in X.
This can be summarized in a general guiding principle: symmetries of the worldsheet theory
correspond to gluing data on the target space.
Remark 2.1. The action (26) is not quite the most general one that could be taken. In particular,
there is no demand that g be a symmetric tensor, only that it is of type (0, 2). We could, in
principle, replace g with a sum g + B where B ∈ A2 (X) is a two-form on X. This term is called
the B-field, and setting a nonzero value for the B-field has many interesting ramifications.
For this paper, we will focus on the case B = 0. However, even in this case the ability to shift
the B-field to a nonzero value will play a role.
2.2. T-Duality of Strings on Tori. We have observed that the worldsheet theory of the string
enjoys many symmetries. Some of these symmetries correspond to symmetries of the underlying
worldsheet, like the Diff 0 (Σ) and Weyl group symmetries as well as the worldsheet supersym-
metries, while others were symmetries corresponding to the fields themselves, such as the E(d)-
symmetry of d scalar fields. We now explore a new symmetry enjoyed by both the bosonic string
and the superstring known as T-duality, which is of a different flavor.

2.2.1. Bosonic Strings on the Circle. Consider the bosonic sigma model defined in section 2.1.3,
and set the target to be the circle SR1 of radius R. This is the theory of a single bosonic field X
which is 2πR-periodic in the sense that
X(σ, τ ) = X(σ, τ ) + 2πR (27)
for all (σ, τ ) ∈ Σ.
8 DANIEL M. HALMRAST

Let us revisit the classical solution of (12). There, we argued that PL = PR since a shift of
σ 7→ σ + 2π resulted in a shift of the field to
X(σ, τ ) 7→ X(σ, τ ) + 2πα′ (PL − PR ) (28)
. However, due to the condition (27), PL − PR is no longer zero, but is instead a multiple of αR′ . A
′
solution with nonzero PL − PR is called a winding mode, and the integer α (PLR−PR ) is the winding
number. The target space momentum α′ (PL + PR ) is also constrained since the target manifold is
compact: PL + PR must be an integral multiple of 1/R.
A computation of the mass spectrum [20, Equation 4.15] reveals that a state with PL + PR = Rn
and PL − PR = 2πwR
α′
has total mass

2 n2 w 2 R2
M = 2 + ′2 + oscillator energy (29)
R α
. As the limit R → ∞ is taken, the momentum values become less separated, and approach a
continuum as one would expect of a flat space solution. States of nonzero winding number become
more and more massive, requiring more and more energy to create. In the limit these states freeze
out, and the original solutions to the free boson are recovered.
The limit R → 0 is more interesting. In this limit, the momentum states are frozen out as the
lowest nonzero momentum states gains more and more mass. It is now the winding states that
approach a continuum. Effectively, a new dimension has appeared in which the “winding states”
of the original circle theory become momentum states on the new dimension. Notice this is a
purely “stringy” phenomenon; particles on S 1 can’t wrap the circle, so although their momentum
behaves the same as that of the string there is no winding state that appears. Only in the string
theory do we see the effective new dimension appear.
There is a symmetry here that relates these two modes. To see it, rewrite the solutions of (12)
as
1 
XR (σ − ) = X0 − X̂0 + α′ PR (σ − ) + oscillators
2 (30)
1 
XL (σ + ) = X0 + X̂0 + α′ PL (σ + ) + oscillators
2
where we have introduced the redundant coordinate X̂0 . Now, consider the transformation
XL 7→ XL
(31)
XR 7→ −XR
which sends our solution X(σ, τ ) to
1
X 7→ XL − XR = X̂0 + α′ (PL − PR )τ + ‘a′ (PL + PR )σ + oscillators (32)
2
. This solution is identical to the original solution (14) except X0 has been replaced by X̂0 , and
PL + PR has been replaced by PL − PR . Notice now that the momentum is quantized by values of
R/α′ instead of 1/R. It is a remarkable fact that all aspects of the theory, including the quantum
theory, also remain unchanged.
Theorem 2.2 (T-Duality on a Circle, [20, Section 4.2] or [14, Section 11.2.2]). The (quantum)
nonlinear sigma model on a circle of radius R is equivalent to the nonlinear sigma model on a
circle of radius R′ = α′ /R under the transformation (31).
This symmetry is called T-duality, and is a worldsheet symmetry of the action, formed in terms
of target space quantities, with no interpretation as a symmetry of the target space.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 9

2.2.2. T-duality for Tori. The symmetry of T-duality can be applied to any factor of target space
that is a circle. That is, if the target space is of the form M = X × (SR1 n )n , T-duality can be
applied to each of the n circle factors resulting in a theory on circles of radii Rn′ = α′ /Rn .
This case was originally worked out in [28]. Consider the theory on a torus T n = Rn /2πΛ
(following the conventions of [20, Section 4.5]). The winding modes are parameterized by Λ,
whereas the momentum modes are quantized to live on the dual lattice Λ∗ ⊆ (Rn )∗ . Choosing a
basis ea for Λ and set e∗a as the dual basis for Λ∗ , let X a be the corresponding coordinate function
on the torus, and P a the dual coordinates on (Rn )∗ /Λ∗ . Then, there are classical solutions for any
⃗n = na ∈ Λ∗ and w ⃗ = wa ∈ Λ, similar to (30):
− 1 a 
a
XR (σ ) = X0 − X̂0 + α′ PRa (σ − ) + oscillations
a
2 (33)
a + 1 a a

′ a +
XL (σ ) = X0 + X̂0 + α PL (σ ) + oscillations
2
with
PLa + PRa = na
(34)
PLa − PRa = wa
. T-duality acts by switching factors between the two lattices, which means it naturally acts on
the enlarged space Rn × (Rn )∗ , which we equip with the neutral metric
 
1n 0
G= (35)
0 −1n
which is positive-definite on Rn and negative definite on (Rn )∗ . Our PL and PR naturally live in
the lattice Λ × Λ∗ inside this space. Under the metric G this lattice is self-dual and even.
The full symmetry group of Rn ×(Rn )∗ preserving the metric is O(n, n) the indefinite orthogonal
group, which acts transitively on the set of even self-dual lattices in Rn × (R∗ )n . Since every
nonlinear sigma model into a torus yields an even self-dual lattice of signature (n, n) by this
construction, the space of all torus theories is parameterized by O(n, n).
There is a large subgroup of O(n, n), however, that leaves the theory invariant. The subgroup
O(n) × O(n) acting on the PL and PR factors separately leaves the whole theory invariant. Of
more interest is the discrete subgroup O(n, n; Z) sending the self-dual lattice to itself. The T-
duality transformations of theorem 2.2 form the subgroup switching a generator of Λ with its dual
in Λ∗ . There is also the subgroup SL(n; Z) which acts to preserve both Λ and Λ∗ separately,
corresponding to the discrete group of symmetries of the target space torus. Finally, there is the
subgroup of B-field transformations which shift the background B-field by an integral form.
Remark 2.3. T-duality can be performed in the more general context of a torus fibration over
a base. In this case, the fibers are dualized and the resulting theories remain isomorphic. The
famous conjecture by Strominger, Yau, and Zaslow of SYZ mirror symmetry [35] roughly asserts
that, allowing for torus fibrations with singular fibers, the phenomenon of mirror symmetry can
be explained using T-duality in this way.
2.2.3. Geometric Realizations of T-duality: Double Field Theory. Let X = T n be an n-dimensional
torus, and consider the nonlinear sigma model with target X as in section 2.2.2. Our guiding
principle is that symmetries of the worldsheet theory correspond to gluing data on the target space.
However, the new symmetry we found in section 2.2.2 is distinctly non-geometric in flavor. That
is, the symmetry group O(n, n; Z) does not act on the theory as any subgroup of Diff(Rn ), and
cannot furnish gluing data in the usual way.
One solution to this, originally due to [33][34] and formalized in [15] [16] is to consider a new
type of geometry known as a “T-fold”. Generally, a T-fold is a manifold in which gluing data can
10 DANIEL M. HALMRAST

be taken to lie in the T-duality group as well as the standard diffeomorphism group. Formally,
this means a T-fold is a collection of open sets with a torus fibration over them, and gluing data
between the open sets which is allowed to lie in O(n, n; Z) as well.
Example 2.4 (c.f. [17]). As discussed in section 2.1.3, a well-defined worldsheet theory can be
constructed locally and glued using symmetries of the Lagrangian. Consider a theory which is
locally a nonlinear sigma model with target space a (trivial) T n -fibration over the unit interval.
The interval can be glued at the endpoints to form the base into an S 1 , and now the fibers must
be glued by a symmetry of the theory. As observed in section 2.2.2, all symmetries of the theory
occur as elements of O(n, n). A general element of the Lie algebra o(n, n) can be written in
block-diagonal form  
f Q
N= (36)
K −f T
and gluing data (corresponding to monodromy data around the base) is given by the O(n, n)
element exp(N ) so long as exp(N ) ∈ O(n, n; Z). The three independent components of this
matrix have geometric interpretations:
• The matrix f ∈ sl(n; Z) corresponds to an element of the mapping class group of T n ,
and thus induces a large diffeomorphism of T n . The presence of −f T indicates that the
induced diffeomorphism of the dual torus is simultaneously performed. In this case, the
theory glues to a nonlinear sigma model with target the twisted T n -bundle over S 1 defined
by exp(f ).
• The matrix K induces B-field transformations of the theory. Thus, the target space is
the trivial T n -bundle over the base S 1 , but the B-field does not have a global description.
Equivalently, the dual T n -bundle cannot be globally defined.
• The matrix Q, called the T-fold flux, induces a T -duality transformation on some of the
circle factors. In this case, the target space geometry is a T-fold, and has no global
description as a manifold.
There is a geometric model for T-folds known as doubling spaces, and the corresponding string
theory with T-fold target is known as double field theory.
Remark 2.5. In double field theory, redundant coordinates are added to the action which must be
eliminated by hand using the section constraint. This constraint does not arise as an equation of
motion from the Lagrangian, and because of this the theory is much harder to quantize. Self-dual
Yang-Mills theory also encounters a similar problem in imposing the self-duality conditions on the
field strength tensor [6]. For this reason, we focus on the target space geometric aspects of the
theory and not the worldsheet theory.
Although the specifics of the worldsheet theory lie outside the scope of this paper, the geometric
construction of the target space is the motivation for our main construction. With that in mind,
we review a particular example of a doubling space built in [17].
Example 2.6. Consider the three-dimensional (real) manifold which is a T 2 -fibration over S 1
with monodromy given by the Dehn twist
 
0 0
f= ∈ sl2 (Z)
−m 0
  (37)
f 1 0
e = ∈ SL2 (Z)
−m 1
. In the language of example 2.4, this corresponds to setting f as above with K and Q zero.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 11

Let us coordinatize this manifold in the following way. Let x = x + 1 be a periodic coordinate
parameterizing the base S 1 , and let y = y + 1, z = z + 1 be periodic coordinates on the T 2 fibers
for which the circles y = 0, z = 0 form a basis for the fundamental group. Then, the monodromy
action (37) acts as
x 7→ x + 1
y 7→ y + mz (38)
z 7→ z
and thus the one-forms
P x = dx
P y = dy − mxdx (39)
z
P = dz
are globally defined. We then define the natural metric on this space as g = P x P x + P y P y + P z P z ,
which in matrix form is  
1 0 0
g = 0 1 −mx  (40)
2 2
0 −mx 1 + m x
.
The key idea of double field theory, which is suggested on the worldsheet via the existence of
X̂0 in (30), is to consider the larger space which is a T 2 × T̂ 2 -fibration over the base. That is, the
fibers over the base S 1 are now diffeomorphic to T 4 with coordinates (y, z, ỹ, z̃) and the induced
monodromy is, for x 7→ x + 1,
    
y 1 m 0 0 y
z 
  7→ 0 1 0 0  z 
  
ỹ  0 0 (41)
1 0 ỹ 
z̃ 0 0 −m 1 z̃
. The metric on the original T 2 -fibration extends to a metric on the doubling space by setting the
metric on the fibers to be
 
  1 −mx 0 0
g 0 −mx 1 + m2 x2 0 0 
H= −1 = 
 2 2
 (42)
0 g 0 0 1 + m x mx
0 0 mx 1
. Alongside this metric there is a canonical constant metric L which has signature (2, 2) on the
fibers. In coordinates (y, z, ỹ, z̃):  
0 0 1 0
0 0 0 1
L= 1 0 0
 (43)
0
0 1 0 0
. Notice that the original T 2 and the dual T 2 are both maximal isotropic submanifolds under L.
This doubled manifold can also be viewed as the quotient of a five-dimensional Lie group by a
discrete subgroup. Namely, consider the group G ⊆ GL5 (R) whose general element is of the form
 
1 mx 0 0 y
0 1 0 0 z
 
g(x, y, z, ỹ, z̃) = 0 0
 1 0 ỹ  (44)
0 0 −mx 1 z̃ 
0 0 0 0 1
12 DANIEL M. HALMRAST

with discrete subgroup given by its intersection with GL5 (Z), acting on the left. Explicitly the
integral element g(α, β, γ, β̃, γ̃) acts on the coordinates as
x 7→ x + α
y 7→ y + mαz + β
z 7→ z + γ (45)
ỹ →
7 ỹ + β̃
z̃ →7 z̃ − mαỹ + γ̃
.
To recover the original target space, a projection on the fibers must be specified. Thus, we
define
Definition 2.7. A polarization on a doubled space X is a projection operator
Π ∈ End(T X)
(46)
Π2 = Π
onto a maximally isotropic subbundle of T X. If the distribution defined by Π is integrable, then
the submanifold defined by Π is said to be the physical space-time.
Selecting the projector ΠG to project onto the (y, z) torus recovers the original twisted T 2 -
fibration. However, alternate choices can be made. Consider the projector ΠH defined by pro-
jection onto the (ỹ, z) subspace. By the global description of the monodromy action in (45), this
projector is well-defined on the full T 4 -bundle over S 1 . Let us now consider rewriting everything
in terms of the new splitting of the coordinates into (ỹ, z) and (y, z̃). The metric (42) in these
coordinates is  
1 + m 2 x2 0 0 mx
 0 1 + m2 x2 −mx 0 
H=  (47)
 0 −mx 1 0 
mx 0 0 1
which is not of the form  
g 0
(48)
0 g −1
and thus is not induced by a standard metric on the T 2 -fibration. However, this metric is induced
by a pair of a metric  
1 0
g= (49)
0 1
and a B-field  
0 mx
B= (50)
−mx 0
on the physical T 2 defined by the new projection. Such a target space is known as a T 3 with H-
flux, and is a nontrivial example of a target space with a B-field. In the language of example 2.4,
this corresponds to a nonzero K with zero f and Q.
Of interest to us, however, is the alternate choice of projector ΠT defined by projection to the
(y, z̃) subspace. Reference to the monodromy (45) shows that this projector is only locally defined,
but does not glue to a global operator. Thus, this worldsheet theory, although globally defined on
the worldsheet, lacks an interpretation as a nonlinear sigma model. Such a theory is sometimes
called a nonlinear sigma model with a non-geometric background. In the language of example 2.4,
this corresponds to a nonzero Q with zero K and f .
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 13

The previous example illustrates that the family of worldsheet theories has models with no
geometric interpretation as a nonlinear sigma model. However, by doubling the torus factors a
geometric realization of certain non-geometric sectors can be realized. We will formalize certain
aspects of this construction mathematically in section 4.

2.3. Supersymmetry and the Superstring.

2.3.1. Supersymmetric Strings. Let Σ = R × S 1 again be the worldsheet, coordinatized by (τ, σ)

with σ = σ + 2π. Instead of considering the theory of a scalar function on Σ, as in section 2.1,
we now add to the theory a complex spinor field. Scalar functions (or, more generally, sections
of bundles in the vector representations of Spin(1, 1)) are referred to as bosons, while sections of
spinor bundles are referred to as fermions.
The starting point is the supersymmetric Lagrangian
Z  
µ µ µ 1 2 1 µ¯ µ¯ µ
S [X , ψ− , ψ+ ] = dσ ∂X ∂Xµ + ψ+ ∂ψ+µ + ψ− ∂ψ−µ (51)
4π α′
where we have used the symmetries of Diff 0 (Σ) and the Weyl group to bring the metric γ into
standard form  
1 0
γ= (52)
0 −1
µ µ
. The ψ− are right-moving fermionic fields, and ψ+ are their left-moving counterparts.
This Lagrangian enjoys all the symmetries of the bosonic string, as well as additional symmetries
of the fermionic fields coming from the Spin(1, 1)-representation. However, there is a new type
of symmetry this model enjoys which is qualitatively different than the others. Consider the
symmetry given by
µ µ
X µ 7→ X µ − ϵ− ψ+ + ϵ+ ψ−
µ µ ¯ µ (53)
ψ+ 7→ ψ+ + 2iϵ̄− ∂X µ − 2iϵ̄+ ∂X
where ϵ± are sections of the inverse spinor bundles parameterizing the transformation. This
symmetry mixes bosons and fermions, and symmetries of this type are called supersymmetries.
This theory admitted one supersymmetry dimension for each chirality via ϵ+ and ϵ− , and so this
model is said to be N = (1, 1) supersymmetric.
Remark 2.8. Following the guiding principle of section 2.1.3, one might expect these symmetries
to correspond to a certain type of manifold for which the symmetries are gluing data. This
perspective on supersymmetric sigma models leads to the superspace formalism, where the target
space is a new type of geometric object called a supermanifold. The details of this lie outside the
scope of this paper, but can be found in e.g. [9].
As before, this theory glues to a nonlinear sigma model into an arbitrary Riemannian manifold.
It is shown in [37] and later generalized in [1] that additional supersymmetries can be added with
additional structure on the target manifold. The theory can be upgraded to an N = (2, 2) theory,
where there are two linearly independent transformations of each chirality, provided there exists
a (1, 1) tensor J satisfying
J 2 = −1
(54)
Nij(J) = 0
where Nij is the Nijenhuis tensor. This operator must also be parallel with respect to the metric.
Such an operator defines an integrable almost complex structure on X, and the parallel condition
enforces that the manifold is Kähler.
14 DANIEL M. HALMRAST

2.3.2. The Nonlinear Sigma Model. Let (X, g) now be a Kähler manifold. Denote by capital
letters I, J, etc. indices for the underlying real coordinates on X, and by lowercase letters i, j,
etc. and ī, j̄, etc. indices for the holomorphic (resp. antiholomorphic) coordinates. Let D be the
spin connection on Σ, and R the Riemann tensor on X. Finally, fix square roots of the canonical
bundle on Σ to serve as spinor bundles. With this data, the supersymmetric nonlinear Σ-model
on X is given by Z
1
S [Σ, f, ψ] = (gIJ )∂z xI ∂z xJ
Σ 2
i ī i i ī i (55)
+ giī ψ− Dz ψ− + giī ψ+ Dz ψ+
2 2
i ī j j̄


+ (Riīj j̄ ψ+ ψ+ ψ− ψ− )(idz ∧ dz̄)
where gIJ is the Riemannian metric, and giī is the associated Hermitian metric. The fields are
taking values in
Field Bundle
√ (1,0)
ψ+i
K ⊗ f ∗ (TX )
√ (0,1)
ψ+ī
K ⊗ f ∗ (TX )
√ (1,0)
ψ−i
K ⊗ f ∗ (TX )
√ (0,1)
ψ−ī
K ⊗ f ∗ (TX )
. Here f is generally a map from a Riemann surface Σ in to X with components xI . Note that the
first term is the familiar f ∗ g term from the bosonic theory. The last two terms ensure conformal
invariance and N = (2, 2) supersymmetry.

2.3.3. Topological Twisting. In [36], Witten proposes a modification of the supersymmetric non-
linear sigma model which has remarkably nice properties. These theories are called the topological
A and B models, and they are formed by tensoring the fermion bundles on the worldsheet with
cleverly chosen line bundles. We review the construction now.
√ The nonlinear sigma model can be twisted in two ways by tensoring the fermion bundles with
K, its conjugate, and its dual, yielding the A and B model twisted string theories. The A model
twists the fermions to make them take values in
Field Bundle
(1,0)
i
ψ+ f ∗ (TX )
(0,1)
ī
ψ+ K ⊗ f ∗ (TX )
(1,0)
i
ψ− K ⊗ f ∗ (TX )
(0,1)
ī
ψ− f ∗ (TX )
while the B model takes values in
Field Bundle
(1,0)
i
ψ+ K ⊗ f ∗ (TX )
(0,1)
ī
ψ+ f ∗ (TX )
(1,0)
i
ψ− K ⊗ f ∗ (TX )
(0,1)
ī
ψ− f ∗ (TX )
. Either twist destroys half the supersymmetry, and the resulting twisted theory only has two
supercharges instead of four.
We can examine the B model in more detail: the field
η ī = ψ+
ī ī
+ ψ− (56)
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 15

(0,1)
is a section of f ∗ TX , the field
ρi = ψ+
i i
+ ψ− (57)
is a section of f ∗ ΩX (with (1, 0) and (0, 1) components split as above), and
 
ī ī
θi = giī ψ+ − ψ− (58)
(1,0)
is a section of f ∗ ΩX . Taking products of these allows us to associate to each (0, q) form with
V (1,0) V (1,0)
values in p TX a local operator built from the fields. Namely, for θ ∈ Ωq (X, p TX ) given
locally as
j ...j
θ = hī11...īqp dz̄ ī1 ∧ · · · ∧ dz̄ īq ⊗ ∂j1 ∧ · · · ∧ ∂jp (59)
we can associate the operator Oθ defined locally as
j ...j
Oθ = hī11...īqp η ī1 . . . η īq θj1 . . . θjp (60)

The surviving supersymmetry Q is given by

δxi = 0
δxī = iαη ī
i
δψ+ = −α∂xi
ī j̄ ī k̄
(61)
δψ+ = −iαψ− Γj̄ k̄ ψ+
δψ i = −α∂x¯ i
−
ī j̄ ī k̄
δψ− = −iαψ+ Γj̄ k̄ ψ−
√ −1
where α parameterizes the deformation as a section of K . To say that Q is given by the above
means that Q generates the deformation in the sense that
δW = −i {Q(α), W } (62)
for any operator W .
The operator Q satisfies Q2 = 0 up to the equations of motion for the fields, and furthermore
δOθ = {Q, Oθ } = O∂θ
¯ (63)
V (1,0)
. Thus, Q acts on the space of fields, which are sections of Ωq (X, p TX ), as the ∂¯ operator.
In [36], Witten shows that the topologically twisted theories localize, in the sense that the string
states in the topological theory are computed by Q-cohomology.
Theorem 2.9 ([36, Section 4.2]).
Vp In1,0the topological B-model, the string spectrum is given by the
q
Dolbeault cohomology of Ω (X, TX ).
2.3.4. Boundary Conditions for Open Strings. Let us continue focusing on the twisted B-model.
When the worldsheet Σ = R × [0, π] is the worldsheet of an open string, additional terms can
be added to the action which are integrals of forms supported on the boundary ∂Σ. When the
target space X supports a vector bundle E with connection, the Chan-Paton term can be added
to couple the ends of the string to the vector bundle. This amounts to adding a term to the action
I  
S∂Σ = f ∗ (A) − iη ī Fīj ρj (64)
∂Σ
16 DANIEL M. HALMRAST

where A is the connection 1-form for the vector bundle E, and F is its field strength. The operators
V (1,0)
of the theory now take values in Ωq (X, p TX ) ⊗ E (E). However, constraints on the endpoint
nd
involving the field strength F fix the values of θ in terms of η:

θj = Fj k̄ η k̄ (65)

and so the surviving operators are simply (0, q)-forms on X with values in End(E).
This boundary term is invariant under Q only for certain values of A. By varying (64) with the
transformations in (61), the change in the action is zero when E is a holomorphic vector bundle
and the connection is compatible with the holomorphic structure. The supersymmetry Q acts on
operators by the dbar operator ∂¯E on E, and the spectrum is again computed using cohomology.

Proposition 2.10. The string spectrum of the open string B-model with ambient vector bundle E
is given by bundle-valued Dolbeault cohomology of Ωq (X) ⊗ E (E). nd
Remark 2.11. In the topological A-model, a different supersymmetry is preserved. In order to
preserve this supersymmetry, the endpoints no longer couple to holomorphic vector bundles but
instead couple to coisotropic branes [23]. The open string spectrum has not been fully defined in
the A-model, but in the case of the endpoints coupling to Lagrangian submanifolds it is believed
that the open string spectrum is computed via Lagrangian Intersection Floer Theory, defined in
[10].

In general, the open string can couple to many more objects than just holomorphic vector
bundles. By allowing the endpoints of a string to couple to different objects on each end, the
theory has the structure of a category whose objects are boundary conditions strings can couple to,
and whose Hom-spaces are given by the open string spectra with prescribed boundary conditions
on each end. These objects are called D-branes.
In [29] a geometric description for D-branes preserving the supersymmetry is outlined.

Theorem 2.12 ([29, Section 2.3]). The D-branes in a Calabi-Yau target space preserving B-
type supersymmetry compatible with the B-model twist are exactly the holomorphic submanifolds
supporting holomorphic vector bundles.

Remark 2.13. It is commonly accepted that D-branes should be parameterized by the coherent
sheaves on the target space, not just the vector bundles on submanifolds of the target. This
is somewhat justified in [3] where it is shown that certain deformations of the conformal field
theory allow for D-branes to arise as cokernels of sheaf morphisms between the associated locally
free sheaves. This leap from the category of holomorphic vector bundles to coherent sheaves is
discussed in [2, Section 5.3.3].

In [24] the B-model spectrum was computed between two such D-branes:

Theorem 2.14 ([24]). Let E and F be two coherent sheaves on a Calabi-Yau manifold X. Then,
the topological B-model open string spectrum from E to F is given by the Ext-groups
q
M
HomB (E, F) = Extq (E, F) (66)

This was refined in [3], where the total category of such D-branes was computed:

Theorem 2.15 ([3, Section 2.7]). The category of D-branes in the topological B-model on a Calabi-
Yau manifold X is equivalent to the derived category of coherent sheaves on X.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 17

3. Mathematics Background
We now review the prerequisite mathematics for the main construction of the paper. We concern
ourselves primarily with rank one, space filling B-type D-branes, as defined in section 2.3.4, and
we set our target space to be a complex torus. In this chapter, we will develop the theory of
line bundles with connections, focusing in particular on the construction of differential characters
by Cheeger and Simons, defined in [8]. We specialize these results to line bundles over complex
tori, and review the computation of cohomology for these line bundles. Finally, we review the
mathematics of generalized complex geometry, developed in [12].
For this chapter, if X is a projective complex manifold we denote by Ak the sheaf of smooth dif-
ferential k-forms on X, and by Ap,q the subsheaves of p, q forms given by the Hodge decomposition.
In particular, Ak (X) is the module of global smooth k-forms. All forms are considered to take
complex values, and we denote by AkR , etc. the corresponding subsheaf of real-valued forms. The
sheaf of holomorphic functions is denoted OX , and the sheaves of holomorphic k-forms on X are
denoted by Ωk . If L is a line bundle over X, then denote by Ak (L) the sheaf of differential k-forms
with values in L, and by Ap,q (L) the sheaf of (p, q) forms with values in L. When necessary, the
module of global sections of Ak (L) will be denoted Ak (X, L).

3.1. Connections on Holomorphic Line Bundles. For this section we follow [19]. Let X be
a projective complex manifold and L → X a complex line bundle.

3.1.1. Connections and Structure on Line Bundles.

Definition 3.1. A (smooth) connection on L is a C-linear sheaf morphism
∇ : A0 (L) → A1 (L) (67)
satisfying the Leibniz rule
∇(f s) = df ⊗ s + f ⊗ ∇s (68)
for f a local smooth function on X and s a local smooth section of L.
Proposition 3.2 ([19, Corollary 4.2.4]). The set of all smooth connections on L naturally forms an
affine space over A1 (X, End(L)) ∼
= A1 (X) the space of global one-forms with values in End(L) ∼ =
∞
C (X, C) i.e. global complex one-forms.
Recall that complex line bundles are simply smooth vector bundles with typical fiber a complex
vector space, such that the transition functions can be chosen to be smooth functions with values
in GL1 (C) ∼
= C∗ . In particular, a complex line bundle is given by the data of a 1-cocycle ψ ∈
1 ∗
H (X, C ).
Definition 3.3. A holomorphic structure on a line bundle L is the data of a holomorphic 1-cocycle
∗
ψ ∈ H 1 (X, OX ) which induces the defining cocycle of L in H 1 (X, C∗ ).
Alternatively, holomorphic line bundles are such that the projection map from the total space
of L to the base space X is a holomorphic map.
Since the underlying manifold is complex, the hodge decomposition of forms allows us to de-
compose a connection ∇ into its (1, 0) and (0, 1) parts
∇ = ∇1,0 + ∇0,1
∇1,0 = Π1,0 ◦ ∇ : A0 (L) → A1,0 (L) (69)
∇0,1 = Π0,1 ◦ ∇ : A0 (L) → A0,1 (L)
18 DANIEL M. HALMRAST

where Πp,q is the projector associated to the Hodge decomposition

M
Ak (L) = Ap,q (L) (70)
p+q=k

. Although connections are in general non-canonical, the holomorphic structure on the base X
induces a natural operator on L:
Proposition 3.4. The operator ∂¯ : Ap,q → Ap,q+1 extends naturally to a C-linear operator
∂¯L : Ap,q (L) → Ap,q+1 (L) (71)
. Moreover, this operator still satisfies the Leibniz identity:
∂¯L (f α) = ∂(f
¯ ) ∧ α + f ∂¯L (α) (72)
for all f ∈ A0 and α ∈ Ap,q (L), and is square-zero
∂¯L2 = 0 (73)
.
A C-linear operator satisfying the conditions (72) and (73) we call a dbar operator on L.
The dbar operators on L parameterize the holomorphic structures on the underling complex
line bundle.
Theorem 3.5 ([19, Theorem 2.6.26]). The association of a holomorphic line bundle L with the
canonical dbar operator ∂¯L is a bijective correspondence between the space of holomorphic structures
on the underlying complex line bundle and the space of dbar operators on L satisfying the Leibniz
rule and squaring to zero.
Definition 3.6. A connection ∇ on L is compatible with the holomorphic structure if
∇0,1 = ∂¯L (74)
.
Remark 3.7. This definition should not be confused with that of a holomorphic (or algebraic)
connection on L. Indeed, a holomorphic connection on L is by definition a C-linear sheaf homo-
morphism
D : L → Ω1X ⊗ L (75)
satisfying the holomorphic version of the Leibniz rule
D(f s) = ∂(f ) ⊗ s + f D(s) (76)
. This condition is far more restrictive than being compatible with the holomorphic structure.
For example, only topologically trivial line bundles admit holomorphic connections, but all line
bundles admit connections compatible with a holomorphic structure.
Proposition 3.8 ([19, Corollary 4.2.13]). The space of all connections on L compatible with the
holomorphic structure naturally forms an affine space over A1,0 (X, End(L)) ∼
= A1,0 (X)).
The final structure to consider imposing on a line bundle is the following:
Definition 3.9. An hermitian structure on a complex vector bundle E is the data of a positive-
definite hermitian form on each fiber of E, varying smoothly. In particular, an hermitian structure
on a complex line bundle is the data of a positive smooth function on the base.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 19

Definition 3.10. A connection ∇ on a complex line bundle L equipped with an hermitian struc-
ture h is said to be compatible with the hermitian structure if
d(h(s1 , s2 )) = h(∇s1 , s2 ) + h(s1 , ∇s2 ) (77)
for all (local) sections s1 , s2 of L.
Proposition 3.11 ([19, Corollary 4.2.11]). Let (L, h) be a complex line bundle with an hermitian
structure. The space of connections on L compatible with the hermitian structure naturally forms
an affine space over A1 (X, End(L, h)) where End(L, h) is the subsheaf of End(L) for which local
sections a satisfy
h(as1 , s2 ) + h(s1 , as2 ) = 0 (78)
for all s1 , s2 local sections of L.
In particular, A1 (X, End(L, h)) ∼= A1R (X, iR)).
Theorem 3.12 ([19, Proposition 4.2.14]). For (L, h) a complex line bundle with an hermitian
structure and a choice ∂¯L of holomorphic structure on L, there exists a unique connection ∇
compatible with both the holomorphic and hermitian structures called the Chern connection on L.
If h is the positive real function on X defining the hermitian structure, then the Chern connection
is given locally by
∇ = d + ∂ log h (79)
.
Line bundles with hermitian structures have an alternative description in terms of their structure
group U (1).
Theorem 3.13. There is a one-to-one correspondence between principal U (1)-bundles over X
and line bundles with hermitian structures (L, h). The forward association is given by taking the
associated bundle to the standard representation of U (1) on C, with hermitian structure given by
the standard hermitian form on C. The backward direction is given by taking the unitary frame
bundle of L.
Remark 3.14. There is a notion of a connection on a principal bundle which induces connections
on associated bundles, and the correspondence of theorem 3.13 extends to connections as well. In
many ways the connections on principal bundles are more well-behaved than their vector bundle
counterparts, but this lies outside the scope of this paper. We will see that the theory of differential
characters will suffice.
3.1.2. Curvature and Chern Classes. Let L be a line bundle over X with connection ∇. We have
observed that ∇ itself is not a one-form, but instead sits in an affine space over A1 (X). The square
of ∇, however, behaves more nicely. Recall that ∇ : A0 (L) → A1 (L) can be extended to higher
degrees by the Leibniz rule (68).
Definition 3.15. The curvature of ∇, denoted F∇ , is the C-linear map
F∇ := ∇2 : A0 (L) → A2 (L) (80)
.
Proposition 3.16 ([19, Lemma 4.3.2]). The curvature is A0 -linear, and thus can be thought of
as an element of A2 (X, End(L)) ∼
= A2 (X). Furthermore, F∇ is d-closed.
Remark 3.17 (Local computations). In a small enough open set U , any connection is of the form
∇=d+A (81)
20 DANIEL M. HALMRAST

for some A ∈ A1 (U ). Then,

F∇ = (d + A)2 = dA (82)
. This makes it clear that
dF∇ = d2 A = 0 (83)
.
Since F∇ is d-closed, it determines a cohomology class in H 2 (X, C). This cohomology class is
independent of choice of connection on L. Indeed, any two connections differ by a one-form:
∇2 − ∇1 = A ∈ A1 (X) (84)
and thus the shift of the curvature
F∇2 = F∇1 + dA (85)
is by an exact form, which does not change the cohomology. Since every line bundle admits an
hermitian structure and a complex structure, we see that the curvature F∇ of any connection on L
determines a cohomology class [F∇ ] ∈ H 1,1 (X, R) which only depends on the underlying complex
line bundle. We formalize this with the following:
Definition 3.18. The first Chern class of L, denoted c1 (L), is the cohomology class
i
c1 (L) = [F∇ ] (86)
2π
where F∇ is the curvature of any connection on L.
Definition 3.19 (Alternate, [11, Page 141]). The first Chern class is alternately described in the
following way. Consider the exponential exact sequence
exp ∗
0 Z OX OX 0 (87)
∗
given by the exponential map from the sheaf OX of holomorphic functions to the subsheaf OX of
invertible holomorphic functions. The long exact sequence on homology induces a map
∗
δ2 : H 1 (X, OX ) → H 2 (X, Z) (88)
. Associating a holomorphic line bundle to its class in H 1 (X, OX ) in the sense of definition 3.3
the map δ2 coincides with the first Chern class map. This explains the normalization factor of 2πi :
the form 2πi F∇ is an integral form.
3.2. Holonomy and Differential Characters. Let (L, h) be a complex hermitian line bundle
over X a projective complex manifold, and let ∇ be a connection on L compatible with h. We have
already seen that the topological type of L is determined by the first Chern class as an element of
H 2 (X, Z), but a finer invariant is needed to determine the isomorphism class of L as a complex
line bundle with U (1)-connection. The theory of differential cohomology, first appearing in [8],
provides a natural framework for this classification, which we will review now. We will mainly
follow [4].

3.2.1. Cheeger-Simons Differential Characters. The theory of Cheeger-Simons differential charac-

ters tracks the information contained in principal bundles with connection. In fact, the theory
extends to the so-called principal n-bundles of higher geometry, but for present purposes we focus
on its application to principal U (1)-bundles.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 21

Definition 3.20. Let X be a smooth manifold. Denote by Ck (X), Zk (X) and Bk (X) the abelian
groups of k-chains, k-cycles, and k-boundaries. A map f ∈ HomZ (Zk−1 (X), U (1)) is said to be a
Cheeger-Simons differential character of degree k if, for every Σ ∈ Ck (X),
 Z 
f (∂Σ) = exp 2πi ω (89)
Σ
k
for some ω ∈ A (X). The form ω is uniquely determined by f , and is called the curvature of f .
We denote it by curv(f ).
The set of all Cheeger-Simons differential characters forms a subgroup
Ĥ k (X) ⊆ HomZ (Zk−1 (X), U (1)) (90)
called the Cheeger-Simons differential cohomology group of degree k.
Example 3.21 (k = 1). In degree 1, the Cheeger-Simons differential cohomology group Ĥ 1 (X)
is a subgroup of HomZ (Z0 (X), U (1)), which can be thought of as Hom(X, U (1)), the group of
(set-theoretic) maps from X to U (1).
Let f ∈ Ĥ 1 (X) be such a map. Fixing a point x0 ∈ X and a point x in a small neighborhood
of x0 , let γ be the straight line connecting x0 to x. Then, since f admits a curvature form (89),
 Z 
f (x) = f (x0 ) exp 2πi curv(f ) (91)
γ

showing that f is, in fact, a smooth function.

Conversely, if f ∈ C ∞ (X, U (1)) is a smooth U (1)-valued function on X, pick a local lift f˜ :
X → R to the universal cover, so that locally
f (x) = exp(2πif˜(x)) (92)
. Then, the one-form df˜ is the curvature of f , showing f ∈ Ĥ 1 (X).
Example 3.22 (k = 2). In degree 2, a Cheeger-Simons differential character f ∈ Ĥ 2 (X) is a map
f : Z1 (X) → U (1) (93)
satisfying (89). This map determines a principal U (1)-bundle with connection, as outlined in [5].
Conversely, suppose P → X is a principal U (1)-bundle with connection ∇. Then, the holonomy
mapping
hol∇ : Z1 (X) → U (1) (94)
defines a map f∇ ∈ HomZ (Z1 (X), U (1)). The curvature F∇ of the connection then functions as
the curvature for f , up to normalization:
i
curv(f ) = F∇ (95)
2π
.
The Cheeger-Simons differential cohomology group has four natural maps associated to it.
• The curvature map
curv : Ĥ k (X) → Ak (X) (96)
sending a character to its curvature form has already been defined. For any f ∈ Ĥ k (X),
curv(f ) is closed and has integral periods, hence the image of curv lies in AkZ (X). Thus
we get a map
δ1 : Ĥ k (X) → AkZ (X) (97)
. We call this map the curvature map.
22 DANIEL M. HALMRAST

• Let χ ∈ H k−1 (X, R/Z), and identify it with a function

χ : Hk−1 (X, Z) → U (1) (98)

in the natural way. Then, χ in particular defines an element of Ĥ k (X) with zero curvature.
This association assembles into a group homomorphism

ι1 : H k−1 (X, R/Z) → Ĥ k (X) (99)

which we call the inclusion of flat characters.

• Let f ∈ Ĥ k (X) and pick a lift of f to the universal cover R of U (1):

f˜ : Zk−1 (X) → R (100)

such that f (γ) = exp(2πif˜(γ)). Now, consider the cochain

δ2 (f ) : Ck (X) → Z
(101)
Z
Σ 7→ curv(f ) − f˜(∂Σ)
Σ

. This is a cocycle, and hence defines a cohomology class δ2 (f ) ∈ H k (X, Z). The notation
is justified by noticing that the cohomology class does not depend on choice of lift. This
assembles into a well-defined group homomorphism

δ2 : Ĥ k (X) → H k (X, Z) (102)

which we call the Chern character map.

• Let ω ∈ Ak−1 (X) be a k − 1-form. Integration against cycles induces a map
Z
ω : Zk−1 (X) → R (103)

which exponentiates to
ι2 (ω) : Zk−1 (X) → U (1)
 Z 
(104)
γ 7→ exp 2πi ω
γ

. By Stokes’ theorem, this map satisfies

 Z 
ι2 (ω)(∂Σ) = exp 2πi dω (105)
Σ

expressing ι2 (ω) as a Cheeger-Simons differential character.

If ω is closed and has integral periods, then ι2 (ω) is the trivial map. Thus, ι2 builds a
well-defined map
ι2 : Ak−1 (X)/Ak−1 k
Z (X) → Ĥ (X) (106)

which we call the topological trivialization map.

 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 23

These four maps allow the Cheeger-Simons differential cohomology group to fit into the differ-
ential character diagram:
0 0

δ
H k−1 (Z) H k−1 (R/Z) H k (Z) H k (R/Z)
r ι1 δ2 i

H k−1 (R) Ĥ k H k (R) (107)

ι2 δ1
deR deR
Ak−1 Ak
Ak−1
Z Ak−1 d
AkZ AkZ
Z

0 0
where each diagonal is an exact sequence. Here, r, δ, i form the long exact sequence in cohomology
associated to the short exact sequence
0 Z R R/Z 0 (108)
, d is the exterior derivative, and deR is the induced map from the De Rham theorem.

3.2.2. Hermitian Line Bundles and Differential Characters. The elements of Ĥ 2 (X) are in bijec-
tive correspondence with isomorphism classes of principal U (1)-bundles over X with a connection,
and hence are in bijection with isomorphism classes of complex line bundles over X with an her-
mitian structure and compatible connection. In this context, the four fundamental morphisms of
a differential character are well-known.
Let (L, h) be a complex line bundle with an hermitian structure over X, and ∇ an hermitian
connection on L. Then,
• δ1 (L, ∇) = curv(∇) = 2πi F∇ is the curvature two-form of the connection.
• ι1 associates to a U (1) representation of π1 (X) the corresponding flat line bundle.
• δ2 (L, ∇) = c1 (L) is the first Chern character of L.
• ι2 associates to the differential form ω the topologically trivial U (1)-bundle with connection
∇ = d + ω.
Let Mh,∇ denote the group (under tensor product) of line bundles with an hermitian structure,
equipped with an hermitian connection. Then, let hol : Mh,∇ → Ĥ 2 (X) be the isomorphism given
by taking the holonomy representation.
Proposition 3.23. The isomorphism between the set of hermitian line bundles with connection
and Ĥ 2 (X) given by the holonomy mapping is a group homomorphism.
Proof. The trivial line bundle with trivial connection OX has zero holonomy, as it is the image
of 0 under ι2 . Hence, hol(OX ) = 0. Now, let (L1 , ∇1 ) and (L2 , ∇2 ) be hermitian line bundles
with connections. Their tensor product is another hermitian line bundle with connection, and if
∇1 = d + A1 and ∇2 = d + A2 are local trivializations, then
∇1 ⊗ ∇2 = d + A1 + A2 (109)
locally.
24 DANIEL M. HALMRAST

The parallel transport map is thus given locally as

 Z 
P∇1 ⊗∇2 (γ) = exp 2πi A1 + A2
γ (110)
= P∇1 (γ)P∇2 (γ)
showing that parallel transport along the tensor product is given locally by multiplication of the
parallel transport maps. Since holonomy can be computed locally in this way, we find that the
holonomy is multiplicative. That is,
hol∇1 ⊗∇2 = hol∇1 hol∇2 (111)
and hence, hol is a group homomorphism. □
3.3. Complex Tori, Cohomology, and Line Bundles. This section follows the excellent ex-
position of [27] and [30], except for subsection 3.3.5, which the author could not find a suitable
reference for.
Much of the main results rely on certain nice properties of line bundles on complex tori. Let V
be an n-dimensional complex vector space and Λ a rank 2n lattice in V . The lattice Λ acts on V
by addition, and the resulting quotient space X = V /Λ is called a complex torus. Topologically,
X∼ = T 2n as smooth manifolds, but the complex structure on V descends to a complex structure
on X making it a dimC (X) = n-dimensional complex manifold. The addition rule on V descends
to the quotient, and makes X an complex abelian Lie group. For any x ∈ X, we denote by
tx : X → X the translation map by x, so that tx (y) = x + y for all y ∈ X.
3.3.1. Cohomology of Complex Tori. The quotient map π : V → X exhibits V as the universal
cover of X, and hence π1 (X)(∼= H1 (X, Z)) ∼
= Λ canonically. Furthermore, by the universal coeffi-
cient theorem there is a natural isomorphism H 1 (X, Z) ∼
= Hom(Λ, Z). This determines the entire
integral cohomology by the Kunneth formula:
k
H (X, Z) ∼
^
k
= H 1 (X, Z) (112)
where the isomorphism is given right-to-left by taking the cup product. The previous two results
then imply
k
H k (X, Z) ∼
^
= Hom(Λ, Z) (113)
. The universal coefficient formula also computes the complex cohomology groups as
H k (X, C) ∼
= H k (X, Z) ⊗ C ∼
= Altk (V, C)
R (114)
where AltkR (V, C) is the space of R-multilinear k-forms from V to C.
On Lie groups, there is a certain class of forms which behave well with respect to the group
operation, and will be central in the main construction.
Definition 3.24. A differential form ω ∈ Ak (X) is said to be translation-invariant if
t∗x ω = ω (115)
for all x ∈ X. The submodule of all translation-invariant k-forms is denoted IFk (X).
Theorem 3.25 ([27, Proposition 1.3.5]). For each cohomology class in H k (X, C), there is a unique
translation-invariant differential k-form representing that class. Thus, the de Rham theorem fur-
nishes an isomorphism
H k (X, C) ∼
= IFk (X) (116)
k
between the k-th complex cohomology group and the group IF (X) of translation-invariant k-forms.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 25

Proof. The de Rham isomorphism identifies H k (X, C) with the cohomology of complex-valued
k-forms on X. If λ1 , . . . , λ2n is a basis for Λ and x1 , . . . , x2n are the corresponding real coordinates
on V , the differentials dxi are all translation-invariant one-forms on X and form a basis for
H 1 (X, Z) (and hence a basis for H 1 (X, C)) dual to the chosen basis for Λ ∼ = H1 (X, Z). By taking
wedge products, we find that a basis for H k (X, C) is given by the k-fold wedge products of the
translation-invariant one-forms dxi . Conversely, each translation-invariant k-form must be a real
linear combination of k-fold wedge products of dxi , and we find that each cohomology class in
H k (X, C) has a canonical representative given by a translation-invariant k-form. That is, there is
an isomorphism
H k (X, C) ∼
= IFk (X) (117)
as desired. □
Remark 3.26. If the torus X is equipped with the standard flat metric, then the translation-
invariant forms are exactly the harmonic forms. Thus, theorem 3.25 is a special case of the Hodge
theorem.
Since V –and hence X as well–comes with a complex structure, the cohomology of X admits a
Hodge decomposition.
Set Ω = HomC (V, C) the space of C-linear maps from V to C and set Ω̄ = HomC̄ (V, C) the space
of C-antilinear maps, so that HomR (V, C) ∼
= Ω ⊕ Ω̄. This decomposition induces a decomposition
on the higher cohomology groups as
p q
∼
M ^ ^
k
H (X, C) = Ω ∧ Ω̄ (118)
p+q=k

which coincides with the Hodge decomposition

p q
H q (ΩpX ) ∼
^ ^
= Ω∧ Ω̄ (119)
.
3.3.2. Line Bundles on Complex Tori. For this section, all line bundles are assumed to be holo-
morphic. Let L be a line bundle on X. The pullback bundle π ∗ L along the universal covering
map is a line bundle on V which is necessarily the trivial line bundle V × C. A general fact about
covering spaces is the following:
Proposition 3.27 ([27, Proposition B.1]). Suppose V is the universal cover of X with covering
map π. Then, there is a canonical isomorphism
H 1 (π1 (X), H 0 (OV∗ )) ∼
= ker(π ∗ : H 1 (X, OX
∗
) → H 1 (V, OV∗ )) (120)
between the group cohomology of π1 (X) in H 0 (OV∗ ) and the kernel of the pullback map.
Elements of the group cohomology H 1 (Λ, H 0 (OV∗ )) are called factors of automorphy for the
covering space, and from the above result we see that every line bundle on X is described by a
factor of automorphy.
Explicitly, a 1-cocycle from Λ to H 0 (OV∗ ) is a holomorphic map f : Λ × V → C∗ satisfying the
cocycle condition
f (λ + µ, v) = f (λ, µ + v)f (µ, v) (121)
. This defines an action of Λ on the trivial line bundle V × C by
λ · (v, z) = (v + λ, f (λ, v)z) (122)
for all λ ∈ Λ. The quotient space V × C/Λ is then a line bundle over X.
26 DANIEL M. HALMRAST

For what follows, let f be a factor of automorphy describing the line bundle L on X. Choose a
logarithm
g :Λ×V →C
(123)
f = exp(2πig)
for f and consider the alternating two-form E ∈ Alt2 (V, Z) given by
EL (λ, µ) = g(µ, λ + v) + g(λ, v) − g(λ, µ + v) − g(µ, v) (124)
for any choice of v ∈ V (the definition is independent of v as a result of the cocycle condition on
f ).
Theorem 3.28 ([27, Theorem 2.1.2]). Under the canonical isomorphism
2
H (X, Z) ∼
^
2
= Hom(Λ, Z) (125)
the first Chern class c1 (L) maps to the alternating form EL .
Conversely, every Hermitian form H : V × V → C whose imaginary part takes integral values
on Λ × Λ induces an alternating form E = Im(H) which is the first Chern class of a line bundle
on X.
Using factors of automorphy all line bundles on X can be characterized.
Definition 3.29. A semi-character for a Hermitian form H with imaginary part E taking integral
values on Λ is a map χ : Λ → U (1) satisfying
χ(γ + µ) = χ(γ)χ(µ) exp(πiE(γ, µ)) (126)
.
A pair (H, χ), with H a Hermitian form whose imaginary part takes integral values on Λ and
χ a semi-character for H, defines a cocycle aH,χ ∈ H 1 (Λ, H 0 (OV∗ )) via
π
aH,χ (γ, v) = χ(γ) exp(πH(v, γ) − H(γ, γ)) (127)
2
. This in turn defines a line bundle L(H, χ) associated to the pair. The converse is also true, and
this allows for a complete characterization of Pic(X).
Theorem 3.30 (Appel-Humbert Theorem, [27, Theorem 2.2.3]). Let P (Λ) be the group consisting
of pairs (H, χ) of a Hermitian form H whose imaginary part takes integral values on Λ and a semi-
character χ, with group operation given by
(H1 , χ1 )(H2 , χ2 ) = (H1 + H2 , χ1 χ2 ) (128)
. Then, the association of (H, χ) with a line bundle defined above induces a group isomorphism
between P (Λ) and Pic(X). Furthermore, c1 (L(H, χ)) = H.
In particular, each line bundle L on X uniquely determines a pair (H, χ) in P (Λ), and has a
canonical choice of representative (the canonical factor ) for its class in H 1 (Λ, H 0 (OV∗ )) given by
aH,χ .
Example 3.31. When H is trivial (H = 0), the semi-characters are group homomorphisms from
Λ to U (1). Line bundles with c1 (L) = 0 are topologically trivial, and form rank-one local systems
on X. The (semi)-character χ classifying L can then be thought of as a unitary representation of
π1 (X) ∼
= Λ which specifies the monodromy of this system.
This description of line bundles also behaves nicely with respect to the group operation.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 27

Proposition 3.32 ([27, Lemma 2.3.2]). If v ∈ V is any lift of x ∈ X, then

t∗x L(H, χ) = L(H, χ exp(2πi Im(H(v, −))) (129)
.
This immediately yields the following well-known result:
Theorem 3.33 (Theorem of the square, [27, Theorem 2.3.3]).
t∗ L∼ = t∗ L ⊗ t∗ L ⊗ L−1
x1 +x2 x1 x2 (130)
3.3.3. Topologically Trivial Line Bundles and the Dual Torus. Recall that Pic0 (X) denotes the
group of topologically trivial line bundles on X, with group operation the tensor product. Ac-
cording to the Appel-Humbert theorem (theorem 3.30), when H = 0 there exists an isomorphism
Hom(Λ, U (1)) ∼
= Pic0 (X) (131)
sending a character χ to the line bundle L(0, χ). This implies that topologically Pic0 (X) is a
torus. We can endow it with a natural complex structure in the following way.
Recall as well that Ω = HomC̄ (V, C) is the vector space of C-antilinear maps from V to C. It is
canonically identified with the real dual V̂ = HomR (V, R) by
Ω → V̂
(132)
f 7→ Im(f )
. Thus, there is a canonical nondegenerate pairing
⟨−, −⟩ : Ω × V → R
(133)
⟨φ, v⟩ = Im(ϕ(v))
which induces a map
exp : Ω → Hom(Λ, U (1))
(134)
φ 7→ exp (2πi⟨φ, −⟩)
whose kernel ker(exp) = Λ̂ can be identified with the dual lattice of Λ under ⟨−, −⟩.
Definition 3.34. The dual torus to X, denoted X̂, is the complex torus
X̂ := Ω/Λ̂ (135)
.
The dual torus can be constructed in various other ways as well.
Proposition 3.35 ([27, Theorem 2.4.1]). The exponential map (134) furnishes an isomorphism
exp : X̂ → Hom(Λ, U (1)) ∼
= Pic0 (X) (136)
. Explicitly, a C-antilinear map φ ∈ HomC̄ (V, C) is mapped to the character χφ defined by
λ 7→ exp (2πi⟨φ, λ⟩) (137)
which, by the Appel-Humbert theorem, defines the unique line bundle L(0, χφ ) ∈ Pic0 (X) with
monodromy given by χφ .
Corollary 3.36. There is a natural identification
H 1 (X, U (1)) ∼
= X̂ (138)
from the natural isomorphism
H 1 (X, U (1)) ∼
= Hom(H1 (X, Z), U (1)) = Hom(Λ, U (1)) (139)
28 DANIEL M. HALMRAST

Proposition 3.37 ([27, Theorem 2.5.1]). The dual torus serves as a fine moduli space for topo-
logically trivial line bundles on X. In particular, there exists a universal bundle P over X × X̂
whose restriction to a fiber P|X×[x̂] is isomorphic to the line bundle defined by x̂. Thus, there is a
natural identification
X̂ ∼
= Pic0 (X) (140)
3.3.4. Cohomology of Line Bundles. The Appel-Humbert theorem gives an explicit description of
Pic(X) as a collection of Pic0 (X)-torsors indexed by the Chern class. We can identify the first
Chern class of a holomorphic line bundle L as the imaginary part of an Hermitian form H on
V . Conversely, every Hermitian form whose imaginary part takes integral values on Λ serves as
the first Chern class for some line bundle. Thus, a line bundle is uniquely specified by H and a
semi-character χ for H. The linear algebra of H is closely related to the cohomology of L, which
we examine now, following [27, Section 2-3].
Definition 3.38. For any line bundle L = (H, χ) on X, define the map ϕL to be
ϕL : X → X̂
(141)
x 7→ t∗x L ⊗ L−1
which, by the theorem of the square, is a homomorphism.
In terms of covering spaces, ϕ has an analytic description.
Proposition 3.39 ([27, Lemma 2.4.5]). The map
ϕH : V → Ω
(142)
v 7→ H(v, −)
is an analytic representation of ϕL in the sense that ϕH descends under the quotient to ϕL : X → X̂.
Using this analytic description, some basic properties of ϕL are derived.
Proposition 3.40 ([27, Corollary 2.4.6]). The following holds for all L = (H, χ):
• The map ϕL only depends on H.
• The association L 7→ ϕL is additive:
ϕL1 ⊗L2 = ϕL1 + ϕL2 (143)
The kernel of ϕL is important and can be computed readily. Define
Λ(L) = {v ∈ V | EH (v, Λ) ⊂ Z} (144)
the set of vectors which pair integrally with Λ under EH := Im(H). Then
K(L) := ker(ϕL ) = Λ(L)/Λ (145)
. By elementary linear algebra, if EH is nondegenerate
deg(ϕL ) = |K(L)| = det(EH ) (146)
, and if det(EH ) = 0 then the kernel has positive dimension.
By the elementary divisor theorem, there exists a symplectic basis of Λ with respect to EH , in
the sense that there is a basis λ1 , . . . , λg , µ1 , . . . , µg for which EH takes form
 
0 D
(147)
−D 0
where D = diag(d1 , . . . , dg ) is a diagonal matrix with positive integer entries di satisfying di |di+1 .
The integers di are uniquely determined by EH , and are called the elementary divisors of EH . If
all di > 0, then EH is nondegenerate.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 29

Definition 3.41. The Pfaffian of EH , denoted Pf(EH ), is

Pf(EH ) = det(D) (148)
and if D contains zeroes on the diagonal, the reduced Pfaffian, denoted Pfr(EH ), is the product of
the nonzero entries of D, unless D = 0 in which case Pfr(D) := 1.
Through some lengthy computations detailed in [27, Section 3], the following cohomological
results are arrived at:
Theorem 3.42 ([27, Corollary 3.2.8]). Suppose H is positive-definite. Then,
h0 (L) = Pf(EH ) (149)
Theorem 3.43 ([27, Theorem 3.3.3]). Suppose H is positive-semidefinite. Denote by K(L)0 the
connected component of the identity in K(L). Then,
(
Pfr(EH ), L|K(L)0 is trivial
h0 (L) = (150)
0, else
Theorem 3.44 ([27, Theorem 3.5.5]). Let H have r positive and s negative eigenvalues. Then,
(
g−r−s


Pfr(EH ), s ≤ q ≤ g − r and L|K(L)0 is trivial
hq (L) = q−s
(151)
0, else
Example 3.45. In the case H = 0, r = s = 0 and K(L)0 = X. Then, all cohomology groups of
L vanish unless L is trivial on all of X, i.e. L ∼
= OX . Then,
 
q 0,q g
h (OX ) = h (X) = (152)
q
as expected.
Example 3.46. In the case of r + s = g, the only nontrivial cohomology group is q = s, which
has dimension Pf(EH ). In other words, if L is a line bundle whose corresponding Hermitian form
is nondegenerate, then (
Pf(EH ) q = s
H q (X, L) = (153)
0 else
.
3.3.5. Line Bundles and Differential Characters on Tori. This subsection is believed to be well-
known, but the author could not find a citation for it.
Let L = (H, χ) be a holomorphic line bundle with Appel-Humbert data given by (H, χ), and
let ∇ be a connection on L. Denote by EH = Im(H) ∈ H 2 (X, Z) the corresponding Chern class.
Definition 3.47. Fix a cohomology class ω ∈ H 2 (X, Z). A line bundle with connection (L, ∇) is
said to be ω-flat if c1 (L) = ω and
F∇ := ∇2 ∈ IF 2 (X) (154)
2
where IF (X) is the space of translation-invariant forms on X.
Remark 3.48. In this notation, an EH -flat connection is an invariant connection on a line bundle
L with associated hermitian form H and Chern class c1 (L) = EH .
Proposition 3.49. The space of EH -flat connections on the complex bundle L forms in a natural
way a torsor over Pic0 (X). In particular, there is a unique EH -flat connection on L which is
compatible with the holomorphic structure.
30 DANIEL M. HALMRAST

Proof. Recall that H : V × V → C is an hermitian form on V whose imaginary part EH := Im(H)

is the first Chern class of L under the identification
2
∼
^
2
H (X, Z) = Hom(Λ, Z) (155)
from the Kunneth formula (112). By theorem 3.25, there is a unique translation-invariant form
FH ∈ IF 2 (X) whose cohomology class coincides with EH . A connection with this curvature can
be written locally as
∇=d+A
(156)
dA = FH
which can be solved for A. Thus, EH -flat connections exist.
For E ∈ H 2 (X, Z), associate to it the unique translation-invariant form E ∈ IF 2 (X) whose
cohomology is E. Then, the space of E-flat connections is precisely the preimage of E under
the curvature map. The group H 1 (X, U (1)) acts on this preimage in the following way: for each
character χ ∈ H 1 (X, U (1)), associate to it the unique line bundle (Lχ , ∇χ ) with flat connection
whose monodromy is given by χ. Then, the action of χ on (L, ∇) is given by tensor product with
Lχ . This is, indeed, well-defined:
c1 (Lχ ⊗ L) = c1 (Lχ ) + c1 (L) = c1 (L) = E
(157)
F∇χ ⊗∇ = F∇χ + F∇ = F∇ = E
.
The action is also free and transitive: if (L1 , ∇1 ) and (L2 , ∇2 ) are two line bundles with E-flat
connections, then L3 = L1 ⊗ L−1 2 is a trivial line bundle with flat connection ∇3 = ∇1 ⊗ (−∇2 ),
and
L3 ⊗ L2 = L1 ⊗ L−1 2 ⊗ L2
= L1
(158)
∇3 ⊗ ∇2 = d + A1 + A2 − A2 locally
= ∇1
showing that the action of H 1 (X, U (1)) is transitive. A similar argument shows the action is also
free.
Hence, the space of E-flat connections is a torsor over H 1 (X, U (1)) as desired. □
When E = 0, an E-flat connection is simply a flat connection. The Riemann-Hilbert correspon-
dence then allows us to identify the space of flat connections with the monodromy representations
of π1 (X). This identification extends to E-flat connections for arbitrary integral classes.
Theorem 3.50. The space of E-flat connections is canonically isomorphic to PicE (X) := c−1
1 (E).

Proof. By [30, Proposition 1.5], every holomorphic line bundle L = (H, χ) on X comes equipped
with a natural hermitian metric defined on the covering space V as
h(v) = exp(−πH(v, v)) (159)
and the Chern connection of this hermitian structure is EH -flat.
Let us follow this argument more closely. Recall that V → X is the universal cover of X, and
L is the quotient of the trivial bundle on V by the action of Λ = H1 (X, Z) defined by
 π 
λ · (v, z) = (v + λ, χ(λ) exp πH(v, λ) − H(λ, λ) z) (160)
2
. By (79) the connection is explicitly
∇ = d − πH(dv, v) (161)
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 31

where H(dv, v) denotes (following the notation of [30]) the one-form

P defined as follows: choose
linear coordinates
P on V so that H is diagonalized: H(v, w) = i vi w̄i . Then for v = (z1 , . . . , zn ),
H(dv, v) = i z̄i dzi .
The curvature of this connection is then readily computed:
F∇ = πH(dv, dv) (162)
P
and, noting that H(dv, dv) is purely imaginary, this is equal to πiEH (dv, dv) = πi i dzi ∧ dz̄i as
an invariant two-form.
Thus, the association of a holomorphic line bundle L = (H, χ) to the Chern connection associ-
ated to the canonical hermitian form (159) yields a map
φ : PicE (X) → curv−1 (EH ) ⊆ Ĥ(X) (163)
whose image lies in the space of EH -flat connections, as (162) shows.
Under the natural action of Pic0 (X) on the space of EH -flat connections, defined in proposi-
tion 3.49, this association is equivariant. Indeed, for L0 = (0, χ0 ) ∈ Pic0 (X) and L = (H, χ) ∈
PicE (X),
L0 · L = L0 ⊗ L = (H, χ0 χ) (164)
and this action does not change the hermitian form (159) on the covering space. Thus, φ(L0 · L)
is computed by multiplying the holonomy by χ0 , which implies
φ(L0 · L) = L0 · φ(L) (165)
as desired. Since a G-equivariant map between G-torsors is an isomorphism, the assertion is
proved. □
Differential characters on tori admit a particularly nice description in terms of semi-characters,
using the top-left to bottom-right exact sequence of (107).
Theorem 3.51. Fix a basis γi for Λ, and regard elements of Λ as elements of Z1 (X) by associating
to any primitive λ ∈ Λ the linear subtorus along the vector λ. Fix as well a closed two-form ω on
X with integral periods. For any choice zi ∈ U (1) of values in U (1) for each γi , there is a unique
differential character f ∈ Ĥ 2 (X) with
curv(f ) = ω (166)
and
f (γi ) = zi (167)
for each i.
Proof. Pick any g ∈ Ĥ 2 (X) with curv(g) = ω. Such a g exists by surjectivity of δ1 in (107).
Then, the regular action of H 1 (X, U (1)) ∼= Pic0 (X) on Ĥ 2 (X) leaves the fibers of δ1 invariant. In
particular, let χ ∈ H 1 (X, U (1)) be the character taking values on γi as
zi
χ(γi ) = (168)
g(γi )
. Then, the differential character χ + g satisfies
curv(χ + g) = curv(g) = ω (169)
and
(χ + g)(γi ) = χ(γi )g(γi ) = zi (170)
as desired.
Clearly such a differential character is unique. If f1 , f2 are such that
curv(f1 ) = curv(f2 ) (171)
32 DANIEL M. HALMRAST

and
f1 (γi ) = f2 (γi ) (172)
then their difference satisfies
curv(f1 − f2 ) = 0 (173)
and
(f1 − f2 )(γi ) = 1 (174)
. But since f1 − f2 is flat, it is determined by the value on basis elements. Thus, f1 − f2 = 0 and
f1 = f2 as desired. □
3.4. Generalized Complex Geometry. In [12], Gualtieri builds the theory of generalized com-
plex geometry, which unifies complex geometry and symplectic geometry in the broad context of
A and B type supersymmetry We follow Gualtieri’s exposition for subsections 3.4.1 and 3.4.2.
3.4.1. Basic Constructions. Let X be a smooth manifold of dimension n. Generalized complex
geometry takes place on the bundle T X ⊕ T ∗ X over X. On this bundle, there is a canonical
neutral (signature (n, n)) metric defined by evaluation:
1
⟨X + ξ, Y + η⟩ = (ξ(Y ) + η(X)) (175)
2
and the musical isomorphism from this metric instantiates the canonical isomorphism
T X ⊕ T ∗X ∼= T ∗X ⊕ T X (176)
.
Definition 3.52 ([12, Definition 4.14]). An almost generalized complex structure on X is an
endomorphism J ∈ End(T X ⊕ T ∗ X) satisfying
J 2 = −1 (177)
and
J ∗ = −J (178)
where J ∗ is the adjoint of J with respect to the canonical metric.
The second condition is equivalent to J being orthogonal with respect to the canonical metric,
since
J ∗ J = (−J )J = −(−1) = 1 (179)
.
Example 3.53. Recall that a complex structure on T X is an endomorphism J of T X satisfying
J 2 = −1. Any complex structure on X is an almost generalized complex structure: take
 
−J 0
JJ = (180)
0 J∗
Example 3.54. Any symplectic structure on X is an almost generalized complex structure: if ω
is the symplectic form, then take
0 −ω −1
 
Jω = (181)
ω 0
The adjective almost suggests that there is an additional integrability condition that should
be imposed. In the case of example 3.53 integrability of a complex structure corresponds to
Frobenius integrability of the +i-eigenspace T 1,0 X of X, i.e. [T 1,0 X, T 1,0 X] ⊆ T 1,0 X. In the case
of example 3.54 the “integrability” condition is dω = 0. Both of these are special cases of a general
notion of integrability with respect to a special bracket operation on T X ⊕ T ∗ X.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 33

Definition 3.55 ([12, Section 3.2]). The Courant bracket on T X ⊕ T ∗ X is the skew-symmetric
bracket defined by
1
[X + ξ, Y + η]C = [X, Y ] + LX η − LY ξ − d (ιX η − ιY ξ) (182)
2
where LX is the Lie derivative along X, and ιX is contraction with X.
Definition 3.56 ([12, Definition 4.18]). An almost generalized complex structure J is said to be
integrable, or a generalized complex structure if the +i-eigenbundle of J is involutive under the
Courant bracket.
Proposition 3.57 ([12, Example 4.20-4.21]). An almost generalized complex structure induced by
an almost complex structure, as in example 3.53, is integrable if and only if the complex structure
is integrable.
Similarly, an almost generalized complex structure induced by a nondegenerate two form, as in
example 3.54, is integrable if and only if the two-form is closed.
3.4.2. Generalized Metrics.
Definition 3.58 ([12, Section 6.1]). A generalized metric on X is a positive-definite metric G on
T X ⊕ T ∗ X compatible with the natural inner product. That is, G2 = 1.
Equivalently, a generalized metric is the data of a middle-dimensional subbundle C+ ⊆ T X ⊕
T ∗ X which is positive-definite with respect to the natural inner product. Define C− to be the
orthogonal complement of C + , and we can define G to be
G = ⟨, ⟩|C+ − ⟨, ⟩|C− (183)
.
A generalized metric is said to be compatible with a generalized complex structure J if the
eigenspaces C± are preserved by J . That is, if J is orthogonal with respect to G. In this case,
since G and J commute, the product GJ = J G defines a second almost generalized complex
structure denoted Jω .
Definition 3.59. A pair (G, J ) of a generalized metric and an almost generalized complex struc-
ture J compatible with G is said to be a generalized Kähler structure on X if both J and Jω are
integrable.
In such a case, the endomorphism Jω is called the generalized Kähler form.
These definitions are motivated by the following illuminating example.
Example 3.60 ([12, Example 6.4]). Let (X, g, J, ω) be a Kähler manifold. Then, J and ω define
generalized complex structures JJ and Jω . Their product defines a positive-definite metric G
which, due to the Kähler condition, can be written in block matrix form as
0 g −1
 
G= (184)
g 0
which is the metric induced by g. The pair (G, JJ ) then defines a generalized Kähler structure.
3.4.3. Submanifolds and Generalized Tangent Spaces. Let M be a submanifold of X. Just as T M
is a subbundle of T X, there is a natural subbundle of T X ⊕ T ∗ X associated to M . The first choice
for the generalized tangent space of M would be the subbundle of (T X ⊕ T ∗ X)|M :
T0 M = T M ⊕ Ann(T M ) (185)
which is a maximal isotropic subbundle of T X ⊕ T ∗ X. However, this definition is not stable under
the symmetries of T X ⊕ T ∗ X (as discussed in [12, Section 7.1]). The correct definition is
34 DANIEL M. HALMRAST

Definition 3.61. Let M be a submanifold of X and F a closed two-form on M . Then, the gen-
eralized tangent bundle of (M, F ), denoted T F M (or T M if the context is clear), is the subbundle
of (T X ⊕ T ∗ X)|M given by
T F M = {X + ξ | X ∈ T M, ξ|M = ιX F } (186)
.
As in the case of complex structures, the generalized tangent bundle identifies if a submanifold
is a generalized complex submanifold:
Definition 3.62. Let M be a submanifold of X along with a closed two-form F on M , and let
J be a generalized complex structure on X. The pair (M, F ) is said to be a generalized complex
submanifold if T F M is stable under J .
Example 3.63. Suppose J is a complex structure on X and J is the induced generalized complex
structure, as in example 3.53. Then, a pair (M, F ) is a generalized complex submanifold if and
only if M is a holomorphic submanifold of X and F is purely of type (1, 1).
If F has integral periods, then it is the curvature of a unitary connection on M which induces
an hermitian holomorphic line bundle for which F is the curvature of the Chern connection.
Example 3.64. Suppose ω is a symplectic form on X and J is the induced generalized complex
structure, as in example 3.54. Then, a pair (M, F ) is a generalized complex submanifold if and
only if either
• M is Lagrangian and F = 0, or
• M is coisotropic and F satisfies the conditions of a rank-one coisotropic brane in the sense
of [23].

4. Main Construction
The ideas of double field theory suggest that it is possible to examine D-branes on a torus by
lifting to a double-dimensional space on which the symmetry group is enlarged to include T -duality
transformations. This program was carried out in the topological A-model in [31], where it was
discovered that the Floer cohomology, appropriately modified, computed the A-model open string
spectra.
We propose a similar construction for the topological B-model, using Y. Qin’s construction of
the doubling space. The lift of a space-filling rank one D-brane in the B-model is defined, and
the link between the generalized complex geometry of the base and the geometry of the lift is
established. We show that, in simple cases, intersection theory computes the correct open string
spectra.
4.1. The Doubling Torus. Let V be an n-dimensional complex vector space, Λ a lattice in V ,
and X = V /Λ the corresponding complex torus. We denote by X̂ := Pic0 (X) the dual torus.
The complex structure on X will be denoted by J ∈ End(T X), J 2 = −1, and the dual complex
structure on X̂ will be denoted by −J ∗ . The product X × X̂ supports the Poincaré bundle P
which expresses X̂ as the moduli space of topologically trivial line bundles on X.
For the topological B-model on X, we propose the following definition of the doubling space of
X:
Definition 4.1. The doubling torus 2 of X, denoted X, is the complex manifold
X = X × X̂ (187)
2Also referred to by some as the doubling space or doubled torus.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 35

with the product complex structure, equipped with a closed nondegenerate two-form c1 (P), the
first Chern class of the Poincaré bundle.
Remark 4.2. In [31, Definition 3.1], Y. Qin considers a symplectic torus (X, ω) instead of a
complex torus, and equips the doubling space with the natural symplectic form ω ⊕ −ω −1 . This
reflects the fact that the topological A-model is defined on symplectic manifolds whereas the
topological B-model is defined on complex manifolds.
The two-form on the doubling torus should be thought of as measuring the product structure
of X × X̂. Indeed, if ei are a basis for H 1 (X, Z) with dual basis e∗i for H 1 (X̂, Z), then
X
c1 (P) = ei ∧ e∗i (188)
i
. This is alternatively thought of as the fundamental two-form defining a para-Hermitian structure
on X.
4.2. Lifting Generalized Complex Structures. As noted in definition 4.1, the doubling torus
comes with a natural complex structure inherited from the complex structure on X and the dual
complex structure on X̂. Furthermore, a symplectic structure on X induces an inverse symplectic
structure on X̂, and in this case the doubling torus inherits a symplectic structure as well. As
Y. Qin observes ([31, Remark 3.2]), this doubled symplectic form functions as a different complex
structure on the doubling space. This is reminiscent of the constructions of generalized complex
geometry, and we make that connection explicit now.
4.2.1. The Tangent Bundle of the doubling torus. Let X be the doubling torus of X, equipped
with projections
X
π π̂ (189)

X X̂
and symplectic form
σ = c1 (P) (190)
. From the local description of σ in (188), it is clear that the fibers of π and π̂ are transverse
Lagrangian foliations of X̂.
We now make clear the connection to generalized complex geometry with the following theorem.
Theorem 4.3. The tangent bundle of X is canonically isomorphic (under σ) to the pullback of
the sum of tangent and cotangent bundles of either X or X̂. That is,
TX ∼= π ∗ (T X ⊕ T ∗ X) ∼
= π̂ ∗ (T X̂ ⊕ T ∗ X̂) (191)
as smooth vector bundles.
Proof. Since X = X × X̂ is globally a product manifold, its tangent bundle splits
TX ∼= π ∗ T X ⊕ π̂ ∗ T X̂ (192)
which are, from a different perspective, the tangent bundles of the two Lagrangian foliations
of X. Since both foliations are Lagrangian, contraction with the symplectic form σ induces an
isomorphism between the normal bundles and the cotangent bundles. Applying this to either
foliation results in the isomorphisms
π ∗ T X ⊕ π̂ ∗ T X̂ ∼
= π∗T X ⊕ π∗T ∗X
(193)
(v, v̂) 7→ (v, ιv̂ σ)
36 DANIEL M. HALMRAST

and
π ∗ T X ⊕ π̂ ∗ T X̂ ∼
= π̂ ∗ T ∗ X̂ ⊕ π̂ ∗ T X̂
(194)
(v, v̂) 7→ (ιv σ, v̂)
. In both cases, we have implicitly used the isomorphism
(π ∗ E)∗ ∼
= π ∗ (E ∗ ) (195)
between the dual of the pullback and the pullback of the dual. □

4.2.2. Lifting Generalized Complex Structures. The identification in theorem 4.3 allows for gener-
alized complex geometry on X to be transported to ordinary geometry on X. Many of the natural
constructions of generalized geometry extend in this way, which we examine now.
Definition 4.4. The foliation operator, denoted by F , is the endomorphism of T X ∼
= π ∗ T X ⊕ π̂ ∗ X̂
∗ ∗
defined as taking the value +1 on π T X and −1 on π̂ T X̂.
The foliation operator is alternatively defined by the pair of projectors Π, Π̂ onto the two factors
of T X by
F = Π − Π̂ (196)
and the association goes in reverse. Any operator F with F 2 = 1 defines projectors by
1
Π= (1 + F )
2 (197)
1
Π̂ = (1 − F )
2
. These define transverse foliations on X if the corresponding distributions are Frobenius integrable.
Observe that for any endomorphism O ∈ End(T X) with block-matrix representation
 
A B
O= (198)
C D
composition with F yields
 
A B
FO = (199)
−C −D
. In particular, the symplectic form σ takes block-diagonal form
 
0 σ|X
σ= (200)
−σ|X 0
and post-composition with F yields the symmetric form
 
0 σ|X
Fσ = (201)
σ|X 0
. Recall as well that the canonical neutral metric on T X ⊕ T ∗ X is given by the evaluation map
⟨−, −⟩ : T X ⊕ T ∗ X → R
(202)
⟨X, ω⟩ = ω(X)
.
Proposition 4.5. The canonical neutral metric on T X ⊕ T ∗ X lifts to the metric defined by F σ.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 37

Proof. By direct computation,

F σ(v + v̂)(w + ŵ) = σ(−v + v̂, w + ŵ)
= σ(v̂, w) + σ(ŵ, v) (203)
= ((ιv̂ σ)(w) + (ιŵ σ)(v))
which is, by virtue of the isomorphism (193) exactly the pairing given by the evaluation map. □
Remark 4.6. Coordinatize X via the coordinates defined in (188). In these coordinates, the
canonical neutral metric is thus X
Fσ = ei ⊗ e∗i (204)
i
.
Remark 4.7. The musical isomorphisms between T X and T ∗ X given by the neutral metric express
the natural isomorphism
π ∗ (T X ⊕ T ∗ X) ∼
= π ∗ (T ∗ X ⊕ T X) (205)
.
Proposition 4.8. Let J ∈ End(T X ⊕ T ∗ X) be an almost generalized complex structure on X.
Then, J lifts to an almost complex structure π ∗ J on X.
Proof. This is a direct consequence of theorem 4.3. Namely, let J : T X ⊕ T ∗ X → T X ⊕ T ∗ X be
such that J 2 = −1 and J ∗ = −J under the canonical pairing on T X ⊕ T ∗ X. This induces an
endomorphism of T X ∼ = π ∗ (T X ⊕ T ∗ X) by pullback:
π ∗ (J ) : T X → T X (206)
and by functoriality of π ∗ this satisfies π ∗ (J )2 = −1 as desired. □
Example 4.9 (Lifts of Symplectic Structures, c.f. [31, Remark 3.2]). Let ω be a symplectic form
on X. Then ω induces a generalized complex structure (example 3.54) defined in block matrix
form as
0 ω −1
 
Jω = (207)
−ω 0
. This lifts to a complex structure on X which is a central object of study in [31].
Example 4.10. Let J be a complex structure on X. Then J induces a generalized complex
structure (example 3.53) defined in block matrix form as
 
J 0
JJ = (208)
0 −J ∗
. From the construction of X̂, we see that −J ∗ is the complex structure on X̂ and thus JJ is
simply the induced complex structure on the product.
Example 4.11 (Lifts of Metrics). Suppose we equip X with a Riemannian metric g. Then, g
induces a metric g −1 on X̂ and the pair forms into a Riemannian metric on X. Using the canonical
metric, this induces an endomorphism G ∈ End(T X) defined in block matrix form as
0 g −1
 
G= (209)
g 0
which satisfies the algebraic properties of a generalized metric.
If X is equipped with a B-field B ∈ H 2 (X, R), then the B-shifted generalized metric is given by
−g −1 B g −1
 
B
G = (210)
g − Bg −1 B Bg −1
38 DANIEL M. HALMRAST

which lifts as well to a positive-definite metric on X.

Example 4.12 (Lifts of Kähler Structures). If (g, J, ω) is a Kähler structure on X, the lifts of
G, JJ , and Jω on X define a Kähler structure on X. Indeed, the lift G of g is a positive-definite
metric and by direct computation Jω = GJJ can be written as
Jω = −π ∗ ω + π̂ ∗ ω −1 (211)
which is the Kähler form associated to the product Kähler manifold (X, g, J, −ω)×(X̂, g −1 , −J ∗ , ω −1 ).
Even if there is no metric on X, the doubling space comes with a canonical metric. With respect
to this metric, a construction similar to example 4.12 can be performed.
Let J be the lift to X of a generalized complex structure on X. Then, the form
J ♯ := ⟨J −, −⟩ : T X → T ∗ X (212)
is a nondegenerate two-form on X.
Example 4.13. Let J be the complex structure on X, and J its lift. Since J is invariant under
translation on X, the form ⟨J−, −⟩ is likewise an invariant form on X. Thus this form is closed
and J ♯ defines a symplectic form on X.
4.3. Lifting D-branes to the Doubling Space. In this context, a (rank one) D-brane is a
pair (M, ∇) of smooth submanifold M along with a U (1)-bundle with U (1) connection ∇. If the
submanifold M is holomorphic and the connection is compatible with the holomorphic structure,
(M, ∇) defines a (rank one) B-type D-brane, whereas if M is Lagrangian and ∇ is flat, then (M, ∇)
defines a (rank one) A-type D-brane. B-type (A-type) D-branes are natural boundary conditions
for the topological B-model (A-model).
In [31], a lift of certain rank one A-type D-branes is defined on a symplectic torus, which we
review now. Let X be a symplectic torus, and L a Lagrangian submanifold of X, endowed with
a flat U (1) connection ∇, or a rank-one coisotropic brane in the sense of [23]. Furthermore, fix a
(set-theoretic) map
ξ : H1 (L) → Z2 (213)
satisfying
ξ(γ + γ ′ ) − ξ(γ) − ξ(γ ′ ) = c1 (∇)(γ ∧ γ ′ ) (214)
in Z2 . Then,
Definition 4.14 ([31, Definition 3.4]). The lift of L is defined to be
{(x, x̂) ∈ X × X̂ | x ∈ L, ⟨x̂, γx ⟩ = (−1)ξ(γx ) hol∇ (γx ), ∀γx ∈ π1 (X, x)} (215)
where γx is any linear circle based at x.
Y. Qin then computes the intersection theory of these lifts, and shows that their Floer intersec-
tion cohomology is closely related to the Fukaya category of X itself. We now attempt to carry
this construction over to the B-model side where X is a complex torus.
To that end, we now let (M, ∇) be a D-brane whose submanifold M is a (closed) holomorphic
subtorus, with F∇ = ∇2 denoting the curvature two-form which we require to be translation-
invariant. The embedding of Λ into X defines linear representatives γ for each element in H1 (M, Z).
Definition 4.15. A symmetric semi-character for E ∈ IF 2 (M ) is a (set-theoretic) map
ξ : Λ → Z2 (216)
satisfying
ξ(γ1 + γ2 ) − ξ(γ1 ) − ξ(γ2 ) = E(γ1 ∧ γ2 ) (217)
for all γ1,2 ∈ H1 (M, Z). Here we implicitly use the isomorphism (116).
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 39

Proposition 4.16. For each invariant two-form E, symmetric semi-characters exist, and the
space of symmetric semi-characters is an affine space over H 1 (M, Z2 ).
Proof. Let {si } freely generate H1 (M, Z), and arbitrarily fix values for ξ(si ). Then, formula (217)
determines the value of ξ on all linear combinations of si , hence all of H1 (M, Z). In particular,
since E(si ∧ si ) = 0, ξ(nsi ) = nξ(si ), the value of ξ is computed inductively as
ξ(ai si ) = ai ξ(si ) + ai aj E(si ∧ sj ) (218)
which satisfies the desired identity.
The difference of any two such functions, say ξ1 , ξ2 , satisfies
(ξ1 − ξ2 )(γ1 + γ2 ) = ξ1 (γ1 ) + ξ1 (γ2 ) − ξ2 (γ1 ) − ξ2 (γ2 ) = (ξ1 − ξ2 )(γ1 ) + (ξ1 − ξ2 )(γ2 ) (219)
hence is an element of H 1 (M, Z2 ). Conversely, adding an element σ ∈ H 1 (M, Z2 ) to ξ clearly
results in a new function satisfying the desired identity. □
Proposition 4.17. For any choice of ξ, the map
v 7→ (−1)ξ(v) hol∇ (x + γv ) (220)
is a group homomorphism from H1 (M, Z) to U (1).
Proof. The obstruction to hol∇ being a group homomorphism is measured by the quotient
hol∇ (x + γv1 +v2 )
q(v1 , v2 ) = (221)
hol∇ (x + γv1 ) hol∇ (x + γv2 )
which in general is not 1 if the curvature of ∇ is nonzero. Using the local expression for the
holonomy along a path γ as  Z 
hol∇ (γ) = exp 2πi A (222)
γ
for A a local gauge potential, we see that Stokes’ theorem guarantees
Z !
q(v1 , v2 ) = exp 2πi F∇ (223)
Σ1,2

where Σ1,2 is the triangular region bounded by γv1 +v2 , γ−v2 and γ−v1 in that order.
The map v 7→ (−1)ξ(v) hol∇ (x + γv ) is a group homomorphism when the quotient
Q(v1 , v2 ) = (−1)ξ(v1 +v2 )−ξ(v1 )−ξ(v2 ) q(v1 , v2 ) (224)
is unity. However, since ξ is a semi-character, this quotient evaluates to
Z !
Q(v1 , v2 ) = (−1)F∇ (v1 ,v2 ) exp 2πi F∇
Σ1,2
Z !! (225)
= exp πi F∇ (v1 , v2 ) + 2 F∇
Σ1,2

. Since F∇ is an invariant form, Z

1
F∇ = F∇ (v1 , v2 ) (226)
Σ1,2 2
resulting in
Q(v1 , v2 ) = 1 (227)
as desired. □
40 DANIEL M. HALMRAST

An element x̂ ∈ X̂ of the dual torus naturally provides a group homomorphism H1 (X, Z) → U (1)
in the following way. Choose any lift v̂ ∈ H 1 (X, R) of x̂ to the universal cover, and for any
v ∈ H1 (X, Z) define
x̂(v) = exp(2πiv̂(v)) (228)
1
as an element of U (1). Notice that any two choices of v̂ differ by an element of H (X, Z) the
dual lattice of H1 (X, Z). Hence, shifting v̂ by an element of H 1 (X, Z) shifts the value of v by an
integer, leaving the expression for x̂ unchanged.
Definition 4.18. The doubled lift of (M, ∇) is defined to be the submanifold
M = {(x, x̂) ∈ X × X̂ | x ∈ M, x̂(v) = (−1)ξ(v) hol∇ (x + γv )∀v ∈ H1 (L, Z)} (229)
Remark 4.19. This is only well-defined if ∇ is F∇ -flat, but this is not as restrictive as one
might think. Indeed, every holomorphic line bundle admits a unique c1 (L)-flat connection, and
computations in the topological B-model show that the holomorphic structure of the line bundle
is the only relevant data to compute the open string spectrum.
4.3.1. Lifting Cheeger-Simons Differential Characters. The lift has a different interpretation in
terms of Cheeger-Simons differential characters, which characterize principal U (1)-bundles with
connection. That is,
Proposition 4.20. There is a one-to-one correspondence between space-filling rank-one D-branes
on X and the second differential cohomology group Ĥ 2 (X) of X. Furthermore, the tensor product
of D-branes corresponds to the addition law on differential cohomology.
Proof. This follows immediately from proposition 3.23. □
Definition 4.21. An E-flat line bundle is called a symmetric E-flat line bundle if the holonomy
around elements of Λ (thought of as linear cycles through the origin) is contained in {±1}. Equiv-
alently, an E-flat line bundle is a symmetric E-flat line bundle if the holonomy is given by the
exponential of a symmetric semi-character (definition 4.15).
Proposition 4.22. Fix an invariant 2-form E ∈ IF 2 (X). For each ξ a symmetric semi-character
as in definition 4.15, there exists a unique symmetric E-flat line bundle with holonomy around
elements of Λ given by (−1)ξ .
Proof. The association
λ 7→ (−1)ξ(λ) (230)
defines a map from the linear 1-cycles Λ on X to U (1). This map satisfies the condition (89) on
linear cycles exactly due to the constraint (217) on ξ.
From the proof of theorem 3.51, we see that such a ξ defines a unique differential character
f ∈ Ĥ 2 (X) satisfying
f (λ) = (−1)ξ(λ) (231)
with
curv(f ) = E ∈ IF 2 (X) (232)
E
. Using the isomorphism between E-flat connections and Pic (X), we associate to f a line bundle
with connection (SE , ∇) with the prescribed holonomy. □
Theorem 4.23. Let L be a line bundle, E = c1 (L), and ∇ an E-flat connection on L. Let S−E
be a symmetric −E-flat line bundle (notice the minus sign!) defined in proposition 4.22. Then,
the lift of (L, ∇) is equivalently described as the graph
 
L(L) = Γ x 7→ (S−E ⊗ t∗−x L) ∈ X̂ (233)
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 41

where X̂ ∼
= Pic0 (X) in the natural way.
Example 4.24 (Lifting the structure sheaf). Consider the trivial D-brane given by (X, d) with d
the exterior derivative, acting on the trivial line bundle X × C. The sheaf of holomorphic sections
of this line bundle is OX . This lift plays special significance, so we denote it by OX .
Since the connection is flat, i.e. d2 = 0, and has trivial monodromy by the definition of the lift
we find the lift is
OX := L(OX ) = {(x, x̂) | (−1)ξ(γi ) x̂(γi ) = 1}
(234)
= {(x, x̂) | eπi(ξ(γi )+2v̂(γi ) = 1}
where v̂ ∈ H 1 (X, R). From this, we see that
1
ξ(γi ) = v̂(γi ) (mod Z) (235)
2
so that v̂ is determined by ξ in H 1 (X, R)/H 1 (X, Z). This therefore determines a point x̂ ∈ X̂ and
the lift of OX is
OX = X × { eπiξ }
 
(236)
.
Using the language of differential characters, this line bundle corresponds to the trivial character
f = 1. We still need to choose a Z2 -character ξ, which corresponds to a choice of symmetric flat
line bundle S0 on X. Its square is trivial, hence this corresponds to a choice of 2-torsion point in
X̂. The lift is then
OX = Γ(x 7→ S0 ⊗ t∗−x OX )
= Γ(x 7→ S0 ) (237)
= X × {[S0 ]}
as computed above. That is, the lift is the fiber over the point S0 .
Example 4.25 (Lifting other flat line bundles). Consider now the D-brane given by (X, ∇) where
∇ is another flat connection on the trivial line bundle X × C. Since ∇ is flat, the holonomy of
∇ is given by monodromy, represented by an element α ∈ H 1 (X, U (1)) ∼ = X̂. Denote by L the
corresponding line bundle.
Choose a ξ as in example 4.24, and denote the corresponding symmetric line bundle as S0 .
Notice that since L is flat, it is translation-invariant. The lift of L is then
L(L) = Γ(x 7→ S0 ⊗ t∗−x L)
= Γ(x 7→ S0 ⊗ L) (238)
= X × {[S0 ⊗ L]}
which, in terms of characters, is the fiber over the point (−1)ξ α.
Example 4.26 (Lifts of E-flat Connections). Let (X, ∇) be a space-filling D-brane with F∇ = iπE
for some fixed E ∈ IF 2 (X) closed with integral periods.
Using the isomorphism in theorem 3.50, we associate to ∇ a holomorphic line bundle L with
c1 (L) = E. We choose as well an −E-flat line bundle S−E as in proposition 4.22. Then, the lift is
defined to be
L(L) = Γ(x 7→ S−E ⊗ t∗−x L) (239)
as before.
More can be said about this map. The line bundle S−E ⊗ t∗−x L can be rewritten as
−1
t∗x S−E ⊗ S−E ⊗ (L ⊗ S−E ) (240)
42 DANIEL M. HALMRAST

which, using the morphism (141), is just

ϕS−E (x) + [L ⊗ S−E ] (241)
, a translate of ϕS−E .
4.4. Geometry of Doubled Branes. To simplify discussion, choose once and for all a symmetric
E-flat line bundle SE for each E, in particular choosing OX for E = 0. We take these symmetric
E-flat line bundles as part of the data for the lift. The choice is arbitrary up to a translation by
a 2-torsion point in X̂.
4.4.1. Properties of the Lift. Similar to what was found in [31], the lifted submanifold enjoys many
nice geometric properties. As before, let L be a line bundle on X and denote its first Chern class
by E := c1 (L). Let L(L) denote the lift of L described in theorem 4.23.
Proposition 4.27 (c.f. [31, Proposition 3.6] in the case of A-branes). The tangent bundle of L(L)
is the subbundle of T X defined by
T L(L) = {(v, v̂) ∈ T X ⊕ T X̂ | ιv̂ σ = ιv E} (242)
. That is, under the isomorphism T X ∼ = π ∗ (T X ⊕ T ∗ X) the tangent bundle of L(L) is isomorphic
to the pullback of the generalized tangent bundle of (X, E).
Proof. Using the description in theorem 4.23 of the lift as the graph of the function
ϕ := ϕS−E + [L ⊗ S−E ] (243)
the tangent space can be computed as the graph of the differential dϕ. However, dϕ = dϕS−E and
ϕS−E has a linear analytic representation
ϕ−E : V → ω
(244)
v 7→ H(v, −)
whose derivative at (x, ϕ(x)), since ϕ−E is linear, is evidently
dϕ−E : Tx X → Tϕ(x) X̂ ∼= T ∗X
x
(245)
v 7→ (w 7→ H(v, w))
. Comparing this to the computation in the proof of theorem 3.50 we recover the description of
the tangent space as the graph of
v 7→ ιv E (246)
as desired. □
The construction of proposition 4.27 reveals that many of the properties that the generalized
tangent bundle of X enjoys are lifted to the doubling space. We illustrate some of these now.
Proposition 4.28. The lift L(L) of a line bundle L is a maximal isotropic submanifold of X
under the canonical neutral metric.
Proof. At a point (x, ϕ(x)) of L(L), the tangent space, by (242), consists of vectors of the form
T L(L) = {v + v̂ | ιv̂ σ = ιv E} (247)
and thus if v + v̂ and w + ŵ are two tangent vectors to L(L),
σ(v + v̂, w + ŵ) = σ(v̂, w) + σ(ŵ, v)
= ιv̂ σ(w) + ιŵ σ(v) (248)
= E(v, w) + E(w, v)
□
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 43

Proposition 4.29. Let J be the complex structure on X induced by the complex structure on
X, and let J ♯ be the induced symplectic form as in example 4.13. Then, the lift L(L) is both
J -holomorphic and J ♯ -Lagrangian.
Proof. Since the tangent bundle of L(L) is identified with the generalized tangent bundle of (X, E),
the endomorphism J acts on L(L) as in [12, Example 7.7]. This is J -holomorphic since E is of
type (1, 1) on X.
This property, combined with proposition 4.28 implies that for every v, w ∈ T L(L),
J ♯ (v, w) = ⟨J v, w⟩ = 0 (249)
since J v ∈ T L(L) and T L(L) is isotropic under the metric. Thus, J ♯ vanishes on L(L) and L(L),
being middle-dimensional, is Lagrangian. □
The intersections of the lifts of all line bundles on X can also be computed using the results of
section 3.3.4. Let ML (X) denote the set of Lagrangian subspaces of X, which carries an action of
X and X̂ = Pic0 (X) given by translation.
Proposition 4.30. The map Pic(X) → ML (X × X̂) sending a line bundle to its lift is Pic0 (X)-
equivariant.
Proof. This follows immediately from the definition of the lift. Let L be arbitrary and L0 ∈
Pic0 (X), represented by a point x̂ ∈ X̂. Then, the lift of their tensor product is
L(L0 ⊗ L) = Γ(x 7→ t∗x S−E ⊗ L ⊗ L0 )
= Γ(x 7→ ϕS−E (x) + [L ⊗ S−E ⊗ L0 ]) (250)
= L(L) + [L0 ]
as desired. □
Proposition 4.31. Let L1 , L2 and L be arbitrary line bundles on X. Then, the operation − ⊗ L
leaves the intersection invariant up to a shift by Pic0 (X). That is,
L(L1 ) ∩ L(L2 ) = tx̂ (L(L1 ⊗ L) ∩ L(L2 ⊗ L)) (251)
for some x̂ ∈ X̂, as submanifolds of X.
Proof. Let Ei = c1 (Li ) and E = c1 (L), recalling that S−Ei is the symmetric −Ei -flat line bundle
fixed at the beginning of this section. If (x, x̂) lies in the intersection of L1 := L(L1 ) and L2 :=
L(L2 ), then by definition
x̂ = S−E1 ⊗ t∗−x L1 = S−E2 ⊗ t∗−x L2 (252)
. After tensoring with L ⊗ S−E , equality still holds at x:
S−E1 ⊗ t∗−x L1 ⊗ L ⊗ S−E = S−E2 ⊗ t∗−x L2 ⊗ L ⊗ S−E (253)
but this may no longer be the line bundle x̂, as it has been shifted by the flat line bundle L ⊗
S−E . □
Using these properties, as well as the results of section 3.3.4, the intersections of lifts of arbitrary
line bundles can be computed.
Theorem 4.32. Let L1 and L2 be two holomorphic line bundles with c1 (L1 ) = c1 (L2 ). Then,
their lifts do not intersect unless L1 = L2 .
Proof. This follows immediately from proposition 4.30, since L2 = L0 ⊗ L1 for some L0 ∈ Pic0 (X).
□
44 DANIEL M. HALMRAST

Theorem 4.33. Let L1 and L2 be arbitrary line bundles, with lifts L1 and L2 . Then, their
intersection is either empty or is given by
L1 ∩ L2 = K(L2 ⊗ L−1
1 ) (254)
up to translations in X̂ and X, where K(L2 ⊗ L−1 1 ) is the kernel of the map defined in (141),
thought of as a submanifold of X via the zero section OX .
Proof. Let Ei = c1 (Li ), and recall that SEi is the chosen symmetric Ei -flat line bundle. By
proposition 4.31, up to a shift in X̂ the intersection can equivalently be computed as
L1 ∩ L2 = L(L1 ⊗ L−1 −1
1 ) ∩ L(L2 ⊗ L1 )
(255)
= OX ∩ L(L2 ⊗ L−1
1 )
.
As was noted in example 4.26, the lift of L2 ⊗ L−1
1 is the graph of the map

ϕ := ϕS−E2 +E1 + x̂1 (256)

for
x̂1 = L2 ⊗ L−1


1 ⊗ S−E2 +E1 (257)
. Since OX is the zero section, the intersection to be computed (255) is the kernel of this map,
which is explicitly computed as the set of x ∈ X satisfying
ϕ(x) = −x̂1 (258)
. But since ϕ is a group homomorphism, the set ϕ−1 (x̂1 ) of solutions is either empty or a translate
in X of the kernel of ϕ. □
Remark 4.34. The case of 4.32 is a special case of 4.33 when E2 = −E1 . In this case,
ϕ = ϕOX + L2 ⊗ L−1


1 (259)
and since ϕOX is the zero map, we recover the desired result.
4.4.2. Relation to Physics: Ext-groups. It is expected that the intersection theory of the doubled
lifts of rank-one B-type D-branes is closely related to the corresponding open string spectrum.
Specifically, in [31] it was noted that the complex structure on X induced by the symplectic
structure on X selects a subspace of the intersection Floer cohomology on the lifts which computes
the A-model open string spectra.
We now show that, in certain cases, a similar phenomenon happens to the intersection theory
of lifts of B-type D-branes with respect to the induced complex structure. Namely, the total
Ext-group between the two line bundles can be computed in this way. Recall (theorem 2.14) the
B-model Hom-spaces are computed using the total Ext-group. We denote by HF∗ the Lagrangian
intersection Floer cohomology.
Definition 4.35 (c.f. [31, Definition 5.1]). Let J be the lift of the complex structure, defined
in example 4.10. The J -holomorphic part of the Floer cohomology ring HF∗ (L, L) ∼ = H ∗ (L, C)
∗
for any lift L is the (0, ∗)-part of H (L, C) under the Hodge decomposition with respect to the
complex structure J .
Theorem 4.36. Let L1 and L2 be two holomorphic line bundles with c1 (L1 ) = c1 (L2 ). Then,
HomB (L1 , L2 ) = HF∗J (L(L1 ), L(L2 ) (260)
where HF∗J ∗
is the J -holomorphic part of HF , and J = JJ is the complex structure on X induced
by the complex structure on X (example 4.10).
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 45

Proof. The B-model morphism space is computed via Ext-groups, i.e.

M
HomB (L1 , L2 ) = Extq (L1 , L2 ) (261)
q

which vanishes unless L1 = L2 . By theorem 4.32, the lifts of these two line bundles do not intersect
unless they are equal as well. When they are equal, by tensoring with their inverse we find that
HomB (L1 , L2 ) = Ext∗ (OX , OX ) = H ∗ (OX ) (262)
which we have computed in example 3.45.
Let L = L(L1 ) be the lift. The Floer cohomology of L is well-known to be
HF∗ (L, L) ∼
= H ∗ (L, C) ∼
= H ∗ (X, C) (263)
. As expected, the J-holomorphic part
HF∗J (L, L) ∼
= H 0,∗ (L, C) ∼
= H ∗ (OX ) (264)
computes the desired ext-group. □

5. Conclusions and Outlook

The ideas of T-duality suggest that the geometry arising from string theory on a torus can
be studied using a corresponding theory on the product of a torus and its dual. We have seen
this bear out in the boundary theory of the topological B-model by defining the doubled lift of a
space-filling rank-one E-flat D-brane and studying the properties of this lift.
The geometry of the doubling space was studied, and it was shown to have a canonical neutral
metric and fundamental two-form. Furthermore, we saw that the complex structure on the original
torus lifted to a complex structure on the doubling space which, paired with the neutral metric,
induced a symplectic structure on the doubling space. The lifts of D-branes were shown to be both
holomorphic under the induced complex structure and maximally isotropic under the canonical
metric, making them Lagrangian with respect to the induced symplectic form.
The intersection theory of these Lagrangian lifts was shown, in the case of two line bundles of
the same topological type, to compute the expected Hom-space of the topological B-model. This
agrees with the parallel analysis of [31] in the A-model case. We also showed that the intersection
properties of these Lagrangian lifts generally is described by the kernel of the well-known morphism
of (141).
Because the lift is Pic0 (X)-equivariant up to translations, computations of arbitrary intersections
reduce to intersections of a modified lift with the zero section OX defined in example 4.24. The
topological B-model Hom-space is expected to be given by
Extq (OX , L) ∼
M
HomB (OX , L) = = H q (X, L) (265)
q

. In the case of L being positive-definite, the intersection of its lift L(L) with OX is zero-
dimensional and cohomology of L is concentrated in degree zero. The naı̈ve expectation that
the cardinality of the intersection agree with h0 (L), however, is incorrect. In fact, comparing
theorem 3.44 with (146) we see that

2
#(L(L) ∩ OX ) = h0 (L) (266)
. One possible hint to resolving this lies in [22] where it was shown that the open string spectrum
for a generalized complex brane is computable using Courant algebroid cohomology. One might
expect that this cohomology has a geometric analogue in the doubling space. We leave this
question, however, for future work.
46 DANIEL M. HALMRAST

References
[1] Luis Alvarez-Gaume and Daniel Z. Freedman, Geometrical structure and ultraviolet finiteness in the super-
symmetric sigma model, Commun. Math. Phys. 80 (1981), 443.
[2] Paul S. Aspinwall, Tom Bridgeland, Alastair Craw, Michael R. Douglas, Anton Kapustin, Gregory W. Moore,
Mark Gross, Graeme Segal, Balázs Szendröi, and P. M. H. Wilson, Dirichlet branes and mirror symmetry,
Clay Mathematics Monographs, vol. 4, AMS, Providence, RI, 2009.
[3] Paul S Aspinwall and Albion Lawrence, Derived categories and zero-brane stability, Journal of High Energy
Physics 2001 (2001), no. 08, 004, arXiv:hep-th/0104147.
[4] Christian Bär and Christian Becker, Differential characters, Lecture Notes in Mathematics, vol. 2112, Springer,
Cham, 2014. MR3237728
[5] J. W. Barrett, Holonomy and path structures in general relativity and Yang-Mills theory, Internat. J. Theoret.
Phys. 30 (1991), no. 9, 1171–1215. MR1122025
[6] Dmitriy Belov and Gregory W. Moore, Holographic action for the self-dual field (2006),
arXiv:hep-th/0605038.
[7] Tom Bridgeland, Stability conditions on triangulated categories, Annals of Mathematics (2007), 317–345,
arXiv:math/0212237.
[8] Jeff Cheeger and James Simons, Differential characters and geometric invariants, Geometry and topology
(College Park, Md., 1983/84), 1985, pp. 50–80. MR827262
[9] P. Deligne, P. Etingof, D. S. Freed, L. C. Jeffrey, D. Kazhdan, J. W. Morgan, D. R. Morrison, and Edward
Witten (eds.), Quantum fields and strings: A course for mathematicians. Vol. 1, 2, 1999.
[10] Kenji Fukaya, Yong-Geun Oh, Hiroshi Ohta, and Kaoru Ono, Lagrangian intersection Floer theory: anomaly
and obstruction. Part I, AMS/IP Studies in Advanced Mathematics, vol. 46.1, American Mathematical Society,
Providence, RI; International Press, Somerville, MA, 2009. MR2553465
[11] Phillip Griffiths and Joseph Harris, Principles of algebraic geometry, Wiley Classics Library, John Wiley &
Sons, Inc., New York, 1994. Reprint of the 1978 original. MR1288523
[12] Marco Gualtieri, Generalized complex geometry, Annals of Mathematics. Second Series 174 (2011), no. 1, 75–
123, arXiv:math/0401221.
[13] Daniel Mark Halmrast, Supersymmetric topological sigma models and doubling spaces, 2024. Thesis (Ph.D.)–
University of California, Santa Barbara.
[14] K. Hori, S. Katz, A. Klemm, R. Pandharipande, R. Thomas, C. Vafa, R. Vakil, and E. Zaslow, Mirror symmetry,
Clay Mathematics Monographs, vol. 1, AMS, Providence, USA, 2003.
[15] C. M. Hull, A Geometry for non-geometric string backgrounds, Journal of High Energy Physics 10 (2005), 065,
arXiv:hep-th/0406102.
[16] C M Hull, Doubled geometry and T-folds, Journal of High Energy Physics 07 (2007), 080,
arXiv:hep-th/0605149.
[17] C. M. Hull and R. A. Reid-Edwards, Non-geometric backgrounds, doubled geometry and generalised T-duality,
Journal of High Energy Physics 09 (2009), 014, arXiv:0902.4032.
[18] Chris Hull and Barton Zwiebach, Double field theory, Journal of High Energy Physics 09 (2009), 099,
arXiv:0904.4664.
[19] Daniel Huybrechts, Complex geometry, Universitext, Springer-Verlag, Berlin, 2005. MR2093043
[20] Clifford V. Johnson, D-Branes, Cambridge Monographs on Mathematical Physics, 2003.
[21] Dominic Joyce and Yinan Song, A theory of generalized Donaldson-Thomas invariants (October 2008),
arXiv:0810.5645.
[22] Anton Kapustin and Yi Li, Open-string BRST cohomology for generalized complex branes, Advances in Theo-
retical and Mathematical Physics 9 (2005), no. 4, 559–574, arXiv:hep-th/0501071.
[23] Anton Kapustin and Dmitri Orlov, Remarks on A branes, mirror symmetry, and the Fukaya category, Journal
of Geometry and Physics 48 (2003), no. 1, 84–99, arXiv:hep-th/0109098.
[24] Sheldon Katz and Eric Sharpe, D-branes, open string vertex operators, and ext groups, Advances in Theoretical
and Mathematical Physics 6 (2002), no. 6, 979–1030, arXiv:hep-th/0208104.
[25] Maxim Kontsevich, Homological algebra of mirror symmetry, Proceedings of the International Congress of
Mathematicians, 1995, pp. 120–139.
[26] Maxim Kontsevich and Yan Soibelman, Stability structures, motivic Donaldson-Thomas invariants and cluster
transformations (2008), arXiv:0811.2435.
[27] Herbert Lange and Christina Birkenhake, Complex abelian varieties, Vol. 302, Springer Berlin, Heidelberg,
2013.
 SUPERSYMMETRIC TOPOLOGICAL SIGMA MODELS AND DOUBLING SPACES 47

[28] K. S. Narain, New heterotic string theories in uncompactified dimensions < 10, Phys. Lett. B 169 (1986),
41–46.
[29] Hirosi Ooguri, Yaron Oz, and Zheng Yin, D-branes on Calabi-Yau spaces and their mirrors, Nucl. Phys. B
477 (1996), 407–430, arXiv:hep-th/9606112.
[30] Alexander Polishchuk, Abelian varieties, theta functions and the Fourier transform, Cambridge Tracts in
Mathematics, vol. 153, Cambridge University Press, Cambridge, 2003. MR1987784
[31] Yingdi Qin, Coisotropic branes on tori and homological mirror symmetry, ProQuest LLC, Ann Arbor, MI,
2020. Thesis (Ph.D.)–University of California, Berkeley. MR4197590
[32] Eric R. Sharpe, D-branes, derived categories, and Grothendieck groups, Nuclear Physics B 561 (1999), 433–450,
arXiv:hep-th/9902116.
[33] W. Siegel, Superspace duality in low-energy superstrings, Phys. Rev. D 48 (1993), 2826–2837,
arXiv:hep-th/9305073.
[34] , Two vierbein formalism for string inspired axionic gravity, Phys. Rev. D 47 (1993), 5453–5459,
arXiv:hep-th/9302036.
[35] Andrew Strominger, Shing-Tung Yau, and Eric Zaslow, Mirror symmetry is T duality, Nucl. Phys. B 479
(1996), 243–259, arXiv:hep-th/9606040.
[36] Edward Witten, Mirror manifolds and topological field theory, AMS/IP Stud. Adv. Math. 9 (1998), 121–160,
arXiv:hep-th/9112056.
[37] B. Zumino, Supersymmetry and Kahler manifolds, Phys. Lett. B 87 (1979), 203.