Fourier-Mukai and Nahm transforms for holomorphic

triples on elliptic curves

Oscar García-Prada¶, Daniel Hernández Ruipérez†, Fabio Pioli¶

Carlos Tejero Prieto§
arXiv:math/0406176v2 [math.AG] 22 Dec 2004

¶ Instituto de Matemáticas y Física Fundamental,

Consejo Superior de Investigaciones Cientı́ficas,
Serrano 113 bis, 28006 Madrid, Spain

† Departamento de Matemáticas and Instituto Universitario de Fı́sica Fundamental y

Matemáticas, Universidad de Salamanca,
Plaza de la Merced 1-4, 37008 Salamanca, Spain

§ Departamento de Matemáticas, Universidad de Salamanca,

Plaza de la Merced 1-4, 37008 Salamanca, Spain

Abstract. We define a Fourier-Mukai transform for a triple consisting of two holomorphic

vector bundles over an elliptic curve and a homomorphism between them. We prove that in
some cases the transform preserves the natural stability condition for a triple. We also define
a Nahm transform for solutions to natural gauge-theoretic equations on a triple — vortices
— and explore some of its basic properties. Our approach combines direct methods with
dimensional reduction techniques, relating triples over a curve with vector bundles over the
product of the curve with the complex projective line.

The authors are members of VBAC (Vector bundles on algebraic curves), which is partially supported by
EAGER (EC FP5 Contract no. HPRN-CT-2000-00099) and by EDGE (EC FP5 Contract no. HPRN-
CT-2000-00101). This research has been partially supported by the Italian/Spain bilateral programme
Azione Integrata, IT203 ”Sheaves on Calabi-Yau manifolds and applications to integrable systems and
string theory” and by the research projects BFM2003-00097 of the spanish DGI and SA118/03 of the
“Junta de Castilla y León”.
E-Mail addresses: [email protected], [email protected], [email protected],
[email protected]
Mathematics Subject Classification: 14D20, 14H60, 14J60 and 14H21.
1 Introduction

The Fourier-Mukai transform, as originally introduced by Mukai for abelian varieties [22]
establishes a duality between the derived categories of coherent sheaves over an abelian
variety and its dual variety. The theory has been extended to more general varieties,
including K3 surfaces, Calabi-Yau threefolds or elliptic fibrations. In particular, it is a
very powerful tool in the study of moduli spaces of vector bundles over abelian surfaces
and K3 surfaces (see [23, 9, 21, 16, 28] for instance). In the gauge-theoretic side, the
Nahm transform provides a differential geometric analogue of the Fourier-Mukai transform
relating instantons (or monopoles) on dual manifolds [24, 17]. In many cases, whenever it
makes sense, both transforms are compatible in a suitable way.
In this paper we study Fourier-Mukai and Nahm transforms for holomorphic triples over
an elliptic curve and their corresponding vortex equations. A triple here consists of two
holomorphic vector bundles over the elliptic curve and a homomorphism between them.
The motivation to study this problem is two-fold. On the one hand the Nahm transform
has been successfully applied to find instanton and monopole solutions, which are defined
in real dimensions 4 and 3 respectively. It is then very natural to try to find an analogue
for 2-dimensional vortices. On the other hand, vortices in two dimensions are equivalent to
SU(2)-invariant instantons over the product of the elliptic curve and the Riemann sphere,
where the SU(2) action is given simply by the usual one on the sphere. This suggests a
relative 4-dimensional approach to the problem. In a related context the Nahm transform
has been successfully applied to study doubly periodic instantons and their relationship
with Hitchin’s equations [18, 19].
Here is a description of the paper. In Section 2 we briefly review the Fourier-Mukai
and Nahm transforms for vector bundles over elliptic curves. We recall the preservation of
stability and prove that the constant central curvature condition for a connection (which
on a curve coincides with the Einstein-Hermitian condition) is preserved. Although the
latter seems to be of general knowledge, we have not found it in the literature and hence
include it here since it is relevant for our further study for triples. We follow the approach
given in [13].
 In Section 3, we review first the basic stability theory for triples. An important feature
is that the stability criterium depends on a real parameter which is typically bounded
[7]. We then introduce the Fourier-Mukai transform for triples on elliptic curves and give
two natural approaches for transforming a triple. The first one is based on the absolute
Fourier-Mukai transform acting on the components of the triple. The second approach
is based on a relative Fourier-Mukai transform combined with a dimensional reduction
procedure. We prove that the Fourier-Mukai transform preserves stability of triples for
“small” and “large” values of the stability parameter, providing an isomorphism of moduli
spaces. What happens for other values of the parameter remains to be investigated. We
conclude this section by applying these results to obtain isomorphisms between moduli
spaces of stable SU(2)-equivariant vector bundles.
Finally, in Section 4, in parallel with Section 3, we develop the formalism for a relative
Nahm transform in the same context. We apply this formalism to transform a solution
to the vortex equations over a triple, regarded as an SU(2)-invariant Einstein-Hermitian
connection on a certain SU(2)-equivariant bundle over the product of the curve with the
complex projective line. In general it seems very hard to identify the equation satisfied by
the Nahm transform of a vortex solution, which one would expect to be again the vortex
equation on the transformed triple. We content ourselves with analysing in full detail the
case of covariantly constant triples, leaving for a future paper the analysis of the general
case. As a byproduct we prove that polystability of triples may not be preserved by the
Fourier-Mukai transform.
In this paper we work over the field of complex numbers C.

2 Fourier-Mukai and Nahm transforms on elliptic cur-

ves

2.1 Fourier-Mukai transform

b = Pic0 (C) be its dual variety. Although C and C
Let C be an elliptic curve and let C b are
isomorphic it will be convenient to keep a notational distinction between them for the sake
of clarity. Over C × C b we consider the Poincaré bundle P and we denote by πC and π b
C
b As it is customary, P is normalized
the canonical projections onto the factors C and C.
2
 b In [22] Mukai introduced a functor between the bounded
so that it is trivial over {0} × C.
derived categories of coherent sheaves of C and C:b

b
S : D(C) → D(C).

This functor acts as follows

∗
S(E) = RπC,∗
b (πC E ⊗ P),

where E is an object of the derived category and RπC,∗

b denotes the derived functor of πC,∗
b .

We need some standard terminology and notation. As usual, we denote by S i (E) the
sheaf defined by the i-th cohomology of the complex S(E), that is,

S i (E) = Hi (S(E)) .

∗
When E is a single sheaf, S i (E) is the ordinary derived functor Ri πC,∗
b (πC E ⊗ P). A sheaf

E is said to be WITi if S j (E) = 0 for every j 6= i, and E is called ITi if it is WITi and
its unique transform S i (E) is locally-free. Equivalently E is ITi if the cohomology group
b where Cξ = C × {ξ} and Pξ
H j (Cξ , E ⊗ Pξ ) = 0 vanishes for every j 6= i and every ξ ∈ C,
b of the unique Fourier-Mukai
is the restriction of P to Cξ . In this case, the fibre over ξ ∈ C
transform S i (E) is canonically isomorphic to H i (Cξ , E ⊗Pξ ). The Fourier-Mukai transform
S i (E) of a WITi sheaf E will be denoted as usual by E. b When there is no need to specify
the index i we shall simply say that a sheaf is WIT or IT.
One of the most important features of the functor S is that it admits an inverse
b → D(C). That is, there are natural isomorphisms:
Ŝ : D(C)

Ŝ ◦ S ≃ IdD(C)
S ◦ Ŝ ≃ IdD(C)
b .

Moreover Ŝ is explicitly given by

ˆ ) = RπC,∗ (π ∗b (F ) ⊗ P ∨ [1]),
S(F C

where P ∨ is the dual of P and [1] is the shift operator.

Let us recall the following well-known fact whose proof relies on the invertibility prop-
erty of the Fourier-Mukai transform (see [13], also [10] and [9]).
3
Proposition 2.1. If E is a semistable (stable) vector bundle of non-zero degree over an
b is also semistable (stable). Moreover,
elliptic curve C, then E is IT and the transform E
E is IT0 (IT1 ) if and only if deg(E) > 0 (deg(E) < 0). Finally, if E is ITi with Chern
b = ((−1)i d, (−1)i+1 r) = (−1)i (d, −r).
character ch(E) = (r, d) then ch(E)

Remark 2.2. If we take into account that any vector bundle E on an elliptic curve
decomposes uniquely into a direct sum of semistable bundles we conclude that E is IT0
(IT1 ) if and only if all of its components have positive (negative) degree. △

Recall that on an elliptic curve C the moduli space MC (r, d) of S-equivalence classes
of semistable bundles of rank r and degree d is isomorphic to the symmetric product S h C,
where h = (r, d) is the greatest common divisor of r and d. When (r, d) > 1 there are
no stable bundles in MC (r, d). When r and d are coprime, all the semistable bundles are
stable and MC (r, d) is isomorphic to C (see [1] and [27] for details, as well as [9] and [16]).
The Fourier-Mukai transform is well-behaved with respect to families of stable bundles
and therefore induces morphisms between moduli spaces. In the case of ITi semistable
bundles on an elliptic curve, the Fourier-Mukai transform also preserves S-equivalence.
More precisely if E is an ITi semistable bundle on C, then it is immediate to see that
every stable bundle Ek in the graded object Gr(E) = ⊕k Ek with respect to a Jordan-
Hölder filtration is ITi . From this follows that if E and E ′ are S-equivalent ITi bundles,
then the transforms E b and E b ′ remain S-equivalent. Therefore we have.

Corollary 2.3. Let MC (r, d) be the moduli space of semistable bundles of rank r and d 6= 0.
Then, in the ITi case, the Fourier-Mukai transform induces an isomorphism between the
moduli spaces
∼
→ MCb ((−1)i d, (−1)i+1 r).
S : MC (r, d) −

Therefore the Fourier-Mukai transform gives rise to an isomorphism between symmetric

products of elliptic curves.

2.2 Nahm transform

We come now to the definition of the Nahm transform in the context of elliptic curves.
Let C be a complex elliptic curve endowed with a flat metric of unit volume. The
canonical spinor bundle S = Λ0,• T ∗ C of C as a spinc manifold, has a natural splitting
4
S = S + ⊕ S − where
S + = Λ0,0 T ∗ C, S − = Λ0,1 T ∗ C .

We denote the spinorial connection of S by ∇S .

b parametrizes the gauge equivalence classes of Hermitian flat
The dual elliptic curve C
line bundles over C. The Poincaré bundle P introduced in Section 2.1 is endowed with
a unitary connection ∇P , such that the restriction of (P, ∇P ) to the slice Cξ is in the
b Therefore for every ξ ∈ C
equivalence class defined by ξ ∈ C. b we have the Hermitian line
bundle Pξ ≡ P|Cξ → C endowed with the flat unitary connection ∇ξ = ∇|Pξ .
Let us consider a Hermitian vector bundle E → C with a unitary connection ∇. On
the vector bundle E ⊗ Pξ we have the connection ∇ξ = ∇ ⊗ 1 + 1 ⊗ ∇ξ . Therefore we have
the family of coupled Dirac operators

Dξ : Ω0 (C, S + ⊗ E ⊗ Pξ ) → Ω0 (C, S − ⊗ E ⊗ Pξ ).

It follows from the Atiyah-Singer Theorem for families that the difference bundle of the
family of Dirac operators D parametrized by C b is a well defined object Ind(D) in K-
theory which is called the index of D. Moreover, if either one of {Ker Dξ } or {Coker Dξ }
has constant rank, then Ker D and Coker D are vector bundles over C b and one has that

b .
Ind(D) = [Ker D] − [Coker D] ∈ K(C)

Definition 2.4. Let (E, ∇) be a pair formed by a Hermitian vector bundle E over C and
a unitary connection ∇ on E. We say that (E, ∇) is an IT (index Theorem) pair if either
Coker D = 0 or Ker D = 0. In the first case we say that (E, ∇) is an IT0 -pair whereas
in the second we call it an IT1 -pair. The transformed bundle of an ITi -pair is the vector
bundle Eb = (−1)i Ind(D) → C b.

Remark 2.5. From a more formal point of view, the study of the family of Dirac operators
D can be approached via the techniques developed by Bismut in his proofs of the Atiyah-
Singer index Theorem for families [5]. In that framework one has to consider the fibration
πb : C × Cb→C b as a family of spinc manifolds, whose fibres are precisely Cξ . The vector
C
bundle of relative spinors is identified with πC∗ S and we can consider then the coupled
relative Dirac operator

D : Ω0 (πC∗ (S + ⊗ E) ⊗ P) → Ω0 (πC∗ (S − ⊗ E) ⊗ P) ,
5
whose restriction to Cξ is Dξ . △

b is a procedure which transforms Hermitian vector

The Nahm transform from C to C
bundles with unitary connections on C into Hermitian vector bundles with unitary con-
b The main idea relies on the fact that the index (minus the index) of the
nections on C.
family D is a finite rank vector bundle whenever Coker D = 0 (Ker D = 0). In certain
cases this is a consequence of a vanishing Theorem of Bochner type. Before doing so we
introduce some more notation and recall the Weitzenböck formula.
b is a holomorphic Hermitian line
We recall that the Poincaré line bundle P → C × C
bundle and that the unitary connection ∇P is compatible with the holomorphic structure.
It is also known that a Hermitian vector bundle E → C with a unitary connection ∇ is
naturally endowed with a holomorphic structure since F ∇ is of type (1, 1) (see [12, 2.1.53]).
Moreover, the spinc Dirac operator Dξ coincides with the Dolbeault-Dirac operator of
E ⊗ Pξ
√
Dξ = 2(∂¯E⊗P
∗
ξ
+ ∂¯E⊗Pξ ),
where ∂¯E⊗Pξ is the Cauchy-Riemann operator of E ⊗ Pξ . Since C is a one dimensional
complex manifold the Dolbeault-Dirac operator Dξ is reduced to

√
Dξ = 2∂¯E⊗Pξ : Ω0 (Cξ , E ⊗ Pξ ) → Ω0,1 (Cξ , E ⊗ Pξ ) .

As a consequence of the Kähler identities (see [12]), the Weitzenböck formula for the
Dirac operator Dξ can be expressed as

Dξ∗ Dξ = 2∂¯E⊗P
∗
∂¯
ξ E⊗Pξ
= ∇∗ξ ∇ξ − iΛF ∇ ⊗ IdPξ , (1)
where iΛF ∇ is the Hermitian endomorphism of E obtained by contracting iF ∇ with the
Kähler form. We have the following vanishing Theorem.

Theorem 2.6. Let (E, ∇) be a pair formed by a Hermitian vector bundle over C and a
unitary connection.

(i) If iΛF ∇ is non-negative and there exists x ∈ C such that iΛF ∇ (x) > 0 then (E, ∇)
is an IT0 -pair.

(ii) If iΛF ∇ is a non-positive and there exists x ∈ C such that iΛF ∇ (x) < 0 then (E, ∇)
is an IT1 -pair.
6
Proof. Let us suppose that iΛF ∇ < 0. If we apply the Weitzenbock formula (1) to a
section s ∈ Γ(C, E ⊗ Pξ ) and we integrate over C we obtain

Z
2
kDξ sk = k∇ξ sk − 2
hiΛF ∇ s, si ω ≥ 0, (2)
C
where ω is the Riemannian volume element of C. From relation (2) we obtain

(a) ∇ s = 0
ξ
Dξ s = 0 ⇐⇒
(b) hiΛF ∇ s, si = 0.

By (a) one sees that hs, si is constant; therefore if there exists x ∈ C such that iΛF ∇ (x) < 0,
then (b) implies that s(x) = 0 and since hs, si is constant, one has s = 0 and (ii) is proved.
By Serre duality we have H 1 (C, E) ≃ H 0 (C, E ∨)∗ , and hence the first statement follows
from the second one.

We can endow the transformed vector bundle of an IT-pair with a Hermitian metric
and a unitary connection in a natural way. This follows from a rather straightforward
application of the theory for families. We briefly recall the main facts of this construction
following the approach of [12, Chapter 3] and [5].
b
Let H±∞ be the space of C ∞ sections of the vector bundle πC∗ (S ± ⊗ E) ⊗ P over C × C.
We may regard H ∞ as the space of C ∞ sections over C b of the infinite dimensional fibre
±
∞ ∞
bundles H± . The fibres H±, ξ are the sets of C ∞ sections over Cξ of πC∗ (S ± ⊗ E) ⊗ P.
Since πC∗ (S ± ⊗ E) ⊗ P is a Hermitian vector bundle, and the fibres Cξ of the projection
b→C
πCb : C × C b carry a natural volume element ω; we can define the Hermitian metric
Z
hh1 , h2 iπCb = hh1 , h2 i ω , (3)
Cξ

∞ 2
on H±, ξ , We then have the Hilbert bundles H± whose fibres H±, ξ are the L -completion
∞
of H±, ξ with respect to this metric.

Let ∇1 be the connection on πC∗ (S ± ⊗ E) ⊗ P obtained from ∇S , ∇ and ∇P . Now we

e on H∞ as follows
define a connection ∇ ±

e D h = ∇1 H h,
∇ b h ∈ H ∞,
for every D ∈ X(C),
D ±

b to C × C.
where D H is the natural lift of the vector field D from C b It is easy to check that
e is a flat connection.
∇
7
 If (E, ∇) is an ITi -pair, then the regularity Theorem for elliptic operators implies that
b is, according to the parity of the index i, a subbundle of H∞ , and hence there is a
E ±
b b induced by
naturally induced metric on E. We also have a natural unitary connection ∇
e and the orthogonal projection P onto E,
the ambient connection ∇ b that is

b = P ◦ ∇.
∇ e

Let us recall that Hodge theory provides an explicit formula for the projector P . Indeed,
b we have
if (E, ∇) is IT0 then for every ξ ∈ C

Pξ = Id −Dξ∗ Gξ Dξ ,

where Gξ is the Green operator of Dξ Dξ∗ . A similar formula holds in the case of an IT1
pair.

b ∇)
Definition 2.7. Let (E, ∇) be an IT-pair. The pair (E, b is called the Nahm transform
of (E, ∇) and is denoted by N (E, ∇).

Remark 2.8. If ∇ and ∇′ are gauge equivalent unitary connections, it follows from the
b and ∇
very definition of the Nahm transform that ∇ c′ are also gauge equivalent unitary
connections.

e
The following is an easy consequence of the flatness of ∇.

b ∇)
Proposition 2.9. Let (E, b be the Nahm transform of an ITi pair. The curvature of ∇
b
is given by
b e ∧ ∇P
e ) ◦ P.
F ∇ = P ◦ (∇P

Moreover, we can express the curvature in terms of the Green operator as follows

b e ∗ ◦ G ∧ ∇D)
e
F ∇ = P ◦ (∇D ◦ P, if E is IT0 .

A similar expression holds in the case of an IT1 pair.

We study now the Nahm transform of a connection with constant central curvature.
Since all the line bundles Pξ are flat they are trivial as smooth bundles and we may
consider the connection ∇P of the Poincaré line bundle as a family of connections ∇ξ on
the trivial line bundle. In the same way if E → C is a Hermitian vector bundle with a
8
unitary connection ∇ then we have a family of connections ∇ξ on E and the family of
Dolbeault-Dirac operators Dξ considered above act now in the same vector bundle
∇ξ
∂ : Ω0 (E) → Ω0,1 (E).

Using a flat holomorphic coordinate z on C and the flat coordinate w which it induces on
b we have
C
∇ξ ∇
∂ = ∂ + π wdz̄ ⊗ IdE ,
b
which clearly shows that this family depends holomorphically on w ∈ C.
The triviality of the holomorphic tangent bundle of C allows to identify Ω0,1 (C) with
Ω0 (C) by contraction with a global anti-holomorphic vector field V . Since the metric on
C is flat, we can choose V such that it is a parallel vector field whose pointwise norm is
equal to 1. We define the operator
∇ξ
Dξ = iV ∂ : Ω0 (E) −→ Ω0 (E) .

Lemma 2.10. The curvature of ∇ξ is related to the operator Dξ by the formula

iΛF ∇ = iΛF ∇ξ = 2[Dξ , Dξ∗ ].

b In
Let us recall that the flat metric of C induces in a natural way a flat metric on C.
the following Theorem we consider unitary connections of constant central curvature on C
and Cb with respect to these metrics.

Theorem 2.11. Let ∇ be a connection on E with constant central curvature with factor
λ ∈ R, that is iΛF ∇ = λ IdE , where λ = 2πµ(E) and µ(E) is the slope of E.

b is a connection on E
1. If deg(E) > 0 then (E, ∇) is an IT0 pair and ∇ b with constant
2π
central curvature with factor λ̂ = − µ(E) .

b is a connection on E
2. If deg(E) < 0 then (E, ∇) is an IT1 pair and ∇ b with constant
2π
central curvature with factor λ̂ = − µ(E) .

Proof. We shall only prove the first case since the second one can be dealt with in a similar
way.
It is well-known ([11] and [25]) that since ∇ has constant central curvature E must be
polystable. The condition deg(E) > 0 implies, due to Proposition 2.1, that (E, ∇) is an
9
 b is a finite
IT0 pair. All the operators Dξ act on Ω0 (E), therefore the bundle of kernels E
rank subbundle of the trivial Hilbert bundle H+ → C b introduced above and P : H+ → E b
is the orthogonal projection. Then we have

b = P ◦ ∇,
∇ e

where ∇e is the natural flat connection on H+ . Taking into account the above identifications,
the curvature of the connection ∇ b of E,
b given in Proposition 2.9, can be expressed as

b e ∗ ◦ Gξ ∧ ∇D
e ξ ) ◦ Pξ ,
F ∇ = Pξ ◦ (∇D ξ (4)

where Gξ is the Green operator of Dξ Dξ∗ .

As we mentioned above, we can choose a flat holomorphic coordinate z on C such that
the Kähler form is expressed as
i
ω = dz ∧ dz̄ .
2
∂
Therefore, locally we may take V = ∂ z̄
. This implies that

Dξ = D0 + πw IdE .

e ξ = πdw ⊗ IdE and ∇D

It is clear now that ∇D e ∗ = πdw̄ ⊗ IdE which upon substitution
ξ
in (4) gives
b
F ∇ = π 2 Pξ ◦ Gξ ◦ Pξ dw̄ ∧ dw,

where we have used the fact that the identity operator commutes with the Green’s operator
Gξ . We then have to prove that for every u ∈ ker Dξ one has

Gξ u = αu + v ,

where α is a constant and v ∈ (ker Dξ )⊥ . To see this suppose that

Gξ u = u ′ + v for u′ ∈ ker Dξ and v ∈ (ker Dξ )⊥ .

Operating by G−1 ∗
ξ = Dξ Dξ we obtain

u = Dξ Dξ∗ u′ + Dξ Dξ∗ v. (5)

But by hypothesis [D0 , D0∗ ] = λ

2
IdE , therefore [Dξ , Dξ∗ ] = λ
2
IdE and, since Dξ u′ = 0,
equation (5) becomes
λ ′
u− u = Dξ Dξ∗ v.
2
10
Now Dξ Dξ∗ v ∈ (ker Dξ )⊥ , since for every u1 ∈ ker Dξ ,
λ
(Dξ Dξ∗ v, u1) = ( v + Dξ∗ Dξ v, u1 )
2
λ
= ( v, u1) + (Dξ v, Dξ u1 )
2
= 0.

Thus u − λ2 u′ ∈ ker Dξ ∩ (ker Dξ )⊥ = {0}. Hence u′ = 2λ−1 u, concluding that

b (2π)2 i (2π)2
F ∇ = −2π 2 λ−1 dw ∧ dw̄ = − dw ∧ dw̄ = − b ⊗ IdEb ,
ω
iλ 2 iλ
b Therefore iΛF ∇b = − 2π Id b as required.
b is the Kähler form of C.
where ω µ(E) E

2.3 Compatibility between the Fourier-Mukai and Nahm trans-

forms, functoriality and invertibility

Let E → C be a Hermitian vector bundle endowed with a unitary connection ∇. As we

have seen the spinc Dirac operator Dξ is identified with the Dolbeault-Dirac operator of
E ⊗ Pξ . Hodge theory and the Dolbeault isomorphism give that

Ker Dξ ≃ H 0 (Cξ , E ⊗ Pξ ) (6)

Coker Dξ ≃ H 1 (Cξ , E ⊗ Pξ ). (7)

If we suppose that E is ITi with respect to the Fourier-Mukai transform S, then the
isomorphisms (6) and (7) mean that (E, ∇) is an ITi -pair with respect to the Nahm
transform. By [2, Theorem 2] or [12, Theorem 3.2.8] we have a natural C ∞ vector bundle
isomorphism induced by Hodge theory

b−∼
φE : E → S i (E).

Moreover, we have the following.

Theorem 2.12. Let E1 , E2 be Hermitian vector bundles over C endowed with unitary
connections ∇1 , ∇2 such that (E1 , ∇1 ), (E2 , ∇2 ) are ITi -pairs with respect to the Nahm
transform. Then we have

b 1, ∇
1. The connections ∇ b 2 are compatible with the holomorphic structures of S i (E1 ),
S i (E2 ), respectively.
11
 2. For every holomorphic morphism Φ : E2 → E1 we have an induced holomorphic mor-
b2 → E
phism N (Φ) : E b1 and a commutative diagram
φE 2
b2
E / S(E2 ) (8)
N (Φ) S(Φ)
 φE 1 
b1
E / S(E1 )

b is a holomorphic line bundle and the connection

Proof. The Poincaré bundle P → C × C
∇P is compatible with the holomorphic structure. This implies that the families of Dirac
b The first statement follows now by a
operators D , D ∗ vary holomorphically with ξ ∈ C.
ξ ξ
standard argument concerning holomorphic families, see [12, Theorem 3.2.8].
Since Φ is holomorphic, the second statement follows immediately in the IT0 case,
because the fibers of the Nahm transforms are given by the kernels of the Cauchy-Riemann
b2,ξ , E
operators. In the IT1 case the fibers E b1,ξ of the Nahm transform at ξ ∈ C
b are given
by the cokernels of the Dirac operators D2,ξ , D1,ξ and we have
∗ ∇∗2,ξ
Coker D2,ξ = Ker D2,ξ = Ker ∂
∗ ∇∗1,ξ
Coker D1,ξ = Ker D1,ξ = Ker ∂
∇∗2,ξ
Now Φ induces a morphism from Ker ∂ to Ω0,1 (C, E1 ⊗ Pξ ) and composing it with the
∇∗1,ξ
orthogonal projection onto Ker ∂ we get a morphism
∇∗2,ξ ∇∗1,ξ
N (Φ)ξ : Ker ∂ → Ker ∂

which by Hodge theory is the unique one that renders commutative the following diagram
φE2 ,ξ
b2,ξ
E / S(E2 )ξ = H 1 (C, E2,ξ )
N (Φ)ξ S(Φ)ξ
 φE1,ξ 
b1,ξ
E / S(E1 )ξ = H 1 (C, E1,ξ )

Since S(Φ) is a vector bundle morphism and φE2 , φE1 are C ∞ vector bundle isomorphisms,
we conclude that N (Φ) is also a C ∞ vector bundle morphism and we have the commutative
diagram (8). Moreover, S(Φ) is a holomorphic morphism and by the first part of the
b 2, ∇
Theorem we have the compatibility between the connections ∇ b 1 and the holomorphic
structures of S(E2 ), S(E1 ), respectively. These facts imply that N (Φ) is a holomorphic
morphism.

12
 If h is an Hermitian metric on a C ∞ vector bundle E then A(E, h) will denote the
space of unitary connections which are compatible with h. On the other hand we will
denote by C(E) the set of holomorphic structures on E. It is well known that there is an
identification
∼
A(E, h) / C(E)
which associates to (E, ∇) the holomorphic vector bundle E = (E, ∂¯∇ ), the inverse cor-
respondence being given by the map which associates to every holomorphic bundle E =
(E, ∂¯∇ ) the unique connection ∇ compatible with the complex structure and the Hermitian
metric. We can rephrase the preceding Theorem by saying that the Nahm transform and
the Fourier-Mukai transform are compatible with this identification. That is to say, the
following diagram is commutative
∼
A(E, h) / C(E)
N S
 
∼
A(E, h) / C(E).

b are in a symmetrical dual relation with one

The curve C and its dual elliptic curve C
another (see [12, Section 3.3.2]). That is, C parametrizes the flat Hermitian line bundles
b therefore Cbb
over C, ≃ C. Moreover, the restriction of the dual of the Poincaré line bundle
bx endowed with the restriction of the connection ∇P ∨ is isomorphic, as
P ∨ to the slice C
a Hermitian line bundle with connection, to the flat Hermitian bundle corresponding to
x. We can hence apply the Nahm construction in order to transform Hermitian vector
b into Hermitian vector bundles with connection over C.
bundles with connection over C
Let ∇ be a connection with constant central curvature different from zero on a bundle
E over C, and let E = (E, ∂¯∇ ) be the corresponding holomorphic vector bundle; then
deg(E) 6= 0. The isomorphisms (6) and (7) imply that E is ITi with respect to the Fourier-
Mukai transform. Let Eb = S i (E) be its unique transform. It is well known, see [22], that
Eb is IT1−i and that there is an isomorphism of holomorphic vector bundles
b
Eb = Sb 1−i (S i (E)) ≃ E. (9)

By Theorem 2.11 ∇b is a constant central curvature connection on E,

b and hence we can
bb b
b By (9) we have an isomorphism
apply to it the Nahm transform to obtain (E, ∇).
bb
E ≃E.
13
 b
b is compatible with the holomorphic structure
Moreover, Theorem 2.12 implies that ∇
of E, and therefore by the results of Donaldson [11], which in particular extend the theorem
of Narasimhan and Seshadri [25] to genus one, we have the following.

Theorem 2.13. If ∇ is a connection with constant central curvature different from zero
b is a connection with constant central curvature on the bundle E,
on E then ∇ b and there
is a natural isomorphism
bb b
b ≃ (E, ∇).
(E, ∇)

Let Ac (E, h) ⊂ A(E, h) be the subspace of constant central curvature connections and
let Cps (E) ⊂ C(E) be the subspace of polystable holomorphic structures on the C ∞ bundle
E. We have the Donaldson–Narasimhan–Seshadri correspondence (the curve version of the
Hitchin–Kobayashi correspondence)
D
Ac (E, h) −−→ Cps (E) .

The content of the preceding Theorem can be summarized by saying that the Nahm
transform and the Fourier-Mukai transform are compatible with the Donaldson–Narasimhan–
Seshadri correspondence. That is, the following diagram is commutative
D
Ac (E, h) / Cps (E)
N S
 
D
b ĥ)
Ac (E, / b .
Cps (E)

These correspondences descend to the quotients by the corresponding gauge groups,

giving a commutative diagram of correspondences between the associated moduli spaces.
First, the Donaldson–Narasimhan–Seshadri correspondence is well known to descend to
moduli spaces, see [20, Chapter VII]. The descent for the Nahm transform follows from
Remark 2.8 and for the Fourier-Mukai transform is a consequence of its functoriality.

3 Fourier-Mukai transforms for holomorphic triples

3.1 Holomorphic triples

A holomorphic triple over a smooth connected curve C is by definition a triple T =

(E1 , E2 , Φ) where Ei , i = 1, 2 are holomorphic vector bundles and Φ ∈ HomC (E2 , E1 ).
14
Let ni and di be the rank and degree of Ei for i = 1, 2. We say that the triple T is of type
(n1 , n2 , d1 , d2 ). There is a notion of stability for triples which depends on a real parameter
α (see [7] for details). The α-degree of T is defined by

degα (T ) = deg(E1 ⊕ E2 ) + n2 α

and the α-slope is by definition

degα (T )
µα (T ) = .
n1 + n2
The stability condition is defined in a similar way as the slope stability for vector bundles,
precisely: T = (E1 , E2 , Φ) is α-stable (resp. α-semistable) if for every non-trivial subtriple
T ′ = (E1′ , E2′ , Φ′ ) we have
µα (T ′) < µα (T ) (resp. ≤).

Here a subtriple means a triple T ′ = (E1′ , E2′ , Φ′ ) and injective homomorphisms γ1 , γ2 of

sheaves such that the following diagram commutes

Φ′
E2′ / E1′
γ2 γ1
 Φ

E2 / E1 .

Most of the properties which are valid for stable bundles carry along to stable triples.
We denote the moduli space of S-equivalence classes of α-semistable triples of type (n1 , n2 ,
d1 , d2 ) by Nα (n1 , n2 , d1 , d2 ) or simply by Nα if there is no need to specify the topological
invariants. Nαs (n1 , n2 , d1 , d2) denotes the moduli space of α-stable triples.
An important feature is that the stability condition gives bounds on the range of the pa-
rameter α. More precisely, if n1 6= n2 and T = (E1 , E2 , Φ) is α-stable of type (n1 , n2 , d1 , d2 )
then necessarily
0 ≤ αm ≤ α ≤ αM
 
where αm = µ1 − µ2 and αM = 1 + |nn11 +n 2
−n2 |
(µ1 − µ2 ) (see [7]). In the case n1 = n2 , α
ranges in [αm , ∞); we will write in this case, αM = ∞. The interval (αm , αM ) is divided
into a finite number of subintervals determined by values of the parameter for which strict
semistability may occur. The stability criteria for two values of α lying between two
consecutive critical values are equivalent (and therefore the corresponding moduli spaces
15
are isomorphic). As in [8] we shall denote by αL the largest critical value, in particular
when αL < α < αM all the moduli spaces Nα are isomorphic.
We use freely the terminology and results of [7]. Corollary 3.6, Proposition 3.17, Corol-
laries 3.19 and 3.20 and Lemma 4.6 of [7] are particularly useful for the understanding of
this paper.
Now we recall how holomorphic triples on an elliptic curve C are related to SU(2)-
equivariant bundles on the elliptic surface C × P1 . In what follows we shall only deal with
SU(2)-equivariant bundles E which admit a C ∞ SU(2)-equivariant decomposition of the
type

E = p∗ E1 ⊕ (p∗ E2 ⊗ q ∗ H ⊗2 ), (10)

where p, q are the canonical projections of C × P1 onto its factors and H is the C ∞ line
bundle over P1 with first Chern number equal to 1.
In the following, if not otherwise stated, an SU(2)-equivariant bundle will always mean
an holomorphic bundle over C × P1 , SU(2)-equivariant, of type given in (10).
We shall need the following formulation of Proposition 2.3 in [7].

Proposition 3.1. Let C be a smooth connected curve, then

(i) There is a one-to-one correspondence between SU(2)-equivariant holomorphic vector

bundles E of type (10) and holomorphic extensions over C × P1 of the form

0 → p∗ E1 → E → p∗ E2 ⊗ q ∗ OP1 (2) → 0

where E1 , E2 are holomorphic vector bundles on C. Here OP1 (2) is the unique line
bundle of degree 2 over P1 .

(ii) There is a (non-unique) functorial correspondence between such extensions and ele-
ments of HomC (E2 , E1 ) and it is given by a functorial isomorphism

σC : Ext1C×P1 (p∗ E2 ⊗ q ∗ OP1 (2), p∗ E1 ) ≃ HomC (E2 , E1 )

∼
induced by the choice of a trace isomorphism tr : H 1 (P1 , OP1 (−2)) −
→ C.

16
Proof. A proof of (i) and (ii) can be found in [14, Proposition 3.9] and [7, Proposition 2.3].
We recall that there is a natural isomorphism (see for instance [15])
∼
→ HomD(C×P1 ) (p∗ E2 ⊗ q ∗ OP1 (2), p∗E1 [1]) .
Ext1C×P1 (p∗ E2 ⊗ q ∗ OP1 (2), p∗ E1 ) − (11)

Now we have
∼
HomD(C×P1 ) (p∗ E2 ⊗ q ∗ OP1 (2), p∗ E1 [1]) −
→ HomD(C×P1 ) (p∗ E2 , p∗ E1 ⊗ q ∗ OP1 (−2)[1])
∼
→ HomD(C) (E2 , Rp∗ (p∗ E1 ⊗ q ∗ OP1 (−2))[1])
−
∼
→ HomD(C) (E2 , E1 ⊗ Rp∗ (q ∗ OP1 (−2))[1])
−
∼
→ HomD(C) (E2 , E1 ⊗C RΓ(P1, OP1 (−2))[1])
−
∼
→ HomD(C) (E2 , E1 ⊗C H 1(P1 , OP1 (−2)))
−

where the second isomorphism is adjunction between direct and inverse images, the third is
the projection formula, the fourth is base-change in the derived category and the last is due
∼
→ H 1 (P1 , OP1 (−2))[−1]
to the fact that since H 0 (P1 , OP1 (−2)) = 0, then RΓ(P1, OP1 (−2)) −
∼
in the derived category. Composition with a trace map tr : H 1 (P1 , OP1 (−2)) −
→ C gives
the isomorphism

σC : Ext1C×P1 (p∗ E2 ⊗ q ∗ OP1 (2), p∗ E1 ) ≃ HomC (E2 , E1 )

of the statement.

Remark 3.2. We can describe quite easily in an explicit form the inverse isomorphism σC−1 .
∼
The inverse of the trace tr−1 : C −
→ H 1 (P1 , OP1 (−2)) defines an element of H 1 (P1 , OP1 (−2))
and via the isomorphism
∼ ∼
→ Ext1P1 (OP1 (2), OP1 ) −
HomD(P1 ) (OP1 (2), OP1 [1]) − → Ext1P1 (OP1 , OP1 (−2))
∼
→ H 1 (P1 , OP1 (−2))
−

induces a morphism tr−1 : OP1 (2) → OP1 [1] in the derived category. Thus, given a mor-
phism Φ : E2 → E1 , one finds that σC−1 (Φ) is the element of Ext1C×P1 (p∗ E2 ⊗q ∗ OP1 (2), p∗E1 )
corresponding to the morphism

p∗ (Φ) ⊗ q ∗ (tr−1 ) : p∗ E2 ⊗ q ∗ OP1 (2) → p∗ E1 [1]

by the isomorphism (11). △

17
Remark 3.3. Two triples (E1 , E2 , Φ) and (E1 , E2 , λΦ) (λ 6= 0) define different extensions
though the same holomorphic bundle. However, they define different SU(2)-equivariant
holomorphic vector bundles (see [13], [7]), because extensions correspond to SU(2)-equivariant
holomorphic vector bundles and not merely to holomorphic vector bundles. △

The correspondence in Proposition 3.1 also preserves stability. Let ωα be the Kähler
α ∗
class over X × P1 defined by ωα = 2
p ωC + q ∗ ωP1 , with α ∈ R+ . The following result is
proved in [7].

Theorem 3.4. Let T = (E1 , E2 , Φ) be an holomorphic triple over a smooth connected curve
C and let E be the holomorphic SU(2)-equivariant bundle C × P1 defined in Proposition
3.1. Then, if E1 and E2 are not isomorphic, T is α-stable if and only if E is slope-stable
with respect to the Kähler form ωα . In the case E1 ≃ E2 then T is α-stable if and only if
Φ 6= 0, E1 ≃ E2 is stable and E decomposes as a direct sum

E ≃ (p∗ E1 ⊗ q ∗ OP1 (1)) ⊕ (p∗ E2 ⊗ q ∗ OP1 (1)) .

Remark 3.5. The proof of statement (ii) in Proposition 3.1 that we have just given above
shows that the correspondence between triples and equivariant bundles also extends to
families. Indeed, families of (stable) triples correspond functorially to families of SU(2)-
equivariant (stable) bundles. This implies that the moduli space Nα of α-stable triples (of a
given topological type) over an elliptic curve corresponds, via the canonical isomorphism of
SU (2)
Proposition 3.1, to a component of the moduli space Mα of SU(2)-equivariant bundles
(defined by the lift of the SU(2) action determined by (10)) stable with respect to the
Kähler form ωα . Therefore we have a canonical identification

∼ (2)
→ MSU
Nα − α .

3.2 Fourier-Mukai transforms for triples

We begin by briefly recalling the main properties of relative Fourier-Mukai transform in

the case of a trivial elliptic fibration over the projective line.
18
 The corresponding functor is then

b × P1 )
SP1 : D(C × P1 ) → D(C (12)
∗ ∗
SP1 (−) = RπC×P
b 1 ,∗ (πC×P1 (−) ⊗ π
C×Cb (P))

where πC×P1 , πC×P

b 1 and πC×C
b 1
b are the canonical projections of C × C × P onto its factors.

As in Section 2.1, this functor is invertible (see for instance [16, 23]).
We also know that the relative Fourier-Mukai transform is compatible with base-change
in the derived category [16]. In particular, for vector bundles E in D(C) and F in D(P1 ) the
base change isomorphism can be described as follows. Let us denote by p̂, q̂ the projections
b × P1 onto its factors. Then
of C
∼
SP1 (p∗ E ⊗ q ∗ F ) −
→ RπC×P
b
∗ ∗ ∗
1 ,∗ (πC×P1 (p E ⊗ q F ) ⊗ π
∗
C×Cb (P))
∼ ∗
−
→ RπC×P
b 1 ,∗ (π b
C×P1
(q̂ ∗ F ) ⊗ πC×
∗ ∗
b (πC E ⊗ P))
C
∼
(13)
∗ ∗ ∗
−
→ RπC×P
b 1 ,∗ (π
C×Cb (πC E ⊗ P)) ⊗ q̂ (F )
∼
→ p̂∗ (RπC,∗
− ∗ ∗ ∗ ∗
b (πC (E) ⊗ P)) ⊗ q̂ (F ) = p̂ (S(E)) ⊗ q̂ (F )

where the second isomorphism is due to q ◦ πC×P1 = q̂ ◦ πC×P

b 1 and p ◦ πC×P1 = πC ◦ πC×
b Cb,

the third is the projection formula and the forth is base change in the derived category.
We also see that given morphisms Φ : E2 → E1 of vector bundles on C and γ : F2 → F1 of
vector bundles on P1 , then the morphism SP1 (p∗ Φ ⊗ q ∗ γ) is identified with p̂∗ (S(Φ)) ⊗ q̂ ∗ γ,
that is, the following diagram is commutative

SP1 (p∗ Φ⊗q ∗ γ)

SP1 (p∗ E2 ⊗ q ∗ F2 ) / SP1 (p∗ E1 ⊗ q ∗ F1 ) (14)
≀ ≀
 p̂∗ (S(Φ))⊗q̂ ∗ γ 
p̂∗ (S(E2 )) ⊗ q̂ ∗ (F2 ) / p̂∗ (S(E1 )) ⊗ q̂ ∗ (F1 )

where the vertical isomorphisms are the base change isomorphisms (13) we have just con-
sidered.
We shall give two natural definitions of the Fourier-Mukai transform of a triple and
show that they are equivalent under the isomorphism given in (ii) of Proposition 3.1. First
we must ensure that the transform of a triple is again a triple.

Definition 3.6. The triple T = (E1 , E2 , Φ) is ITi if both bundles E1 , E2 are ITi with the
same index i.
19
Definition (I) 3.7. Let T = (E1 , E2 , Φ) be an ITi triple. The Fourier-Mukai transform
of T is defined as the triple Tb = (S i (E1 ), S i (E2 ), S i (Φ)). We shall write Tb = (E
b1 , E
b2 , Φ)
b
for the transformed triple.

Since a triple T corresponds exactly to an SU(2)-equivariant bundle E on C × P1 , this

suggests another definition of the Fourier-Mukai transform of an ITi triple as the triple
associated to the transform of the bundle E with respect to the relative transform SP1 . This
observation leads us in a natural way to consider a relative version of the Nahm transform,
an argument that will be pursued in the next section. Note that in order that the relative
transform of the bundle E consists of a single sheaf, we should ensure that E is ITi . This
is achieved by the following Proposition whose proof is a straightforward consequence of
the base change property of the Fourier-Mukai transform.

Proposition 3.8. If E1 and E2 are ITi -bundles with respect to S (with the same index i),
b sits in an exact sequence of the type
then E is ITi with respect to SP1 and its transform E
b1 → E
0 → p̂∗ E b → p̂∗ E
b2 ⊗ q̂ ∗ OP1 (2) → 0.

b is an SU(2)-equivariant bundle on C
Therefore E b × P1 .

Now we can define.

Definition (II) 3.9. We define the Fourier-Mukai transform of a ITi triple T = (E1 , E2 , Φ)
b = S i 1 (E) of the associated SU(2)-equivariant
as the triple associated to the transform E P
1
and ITi bundle E on C × P .

It remains to check that definitions 3.7 and 3.9 are compatible.

Proposition 3.10. Let T be an IT triple and let E be the corresponding invariant bundle
on C × P1 , then the (absolute) Fourier-Mukai transform Tb in Definition 3.7 corresponds
b of Definition 3.9 under the isomorphism given in
to the triple given by the transform E
Proposition 3.1. In other words we have the following commutative diagram
SP1
Ext1C×P1 (p∗ E2 ⊗ q ∗ OP1 (2), p∗ E1 )) / Ext1C×P1 (p̂∗ E b1 ))
b2 ⊗ q̂ ∗ OP1 (2), p̂∗ E
σC ≀ ≀ σCb
 
S
Ext0C (E2 , E1 ) / b2 , E
Ext0C (E b1 )

where the vertical rows are the isomorphisms introduced in Proposition 3.1 and the hori-
zontal isomorphisms are induced from the relative and absolute Fourier-Mukai transforms.
20
Proof. Given a morphism Φ : E2 → E1 , we know by Remark 3.2, that σC−1 (Φ) is the element
of Ext1C×P1 (p∗ E2 ⊗ q ∗ OP1 (2), p∗E1 ) corresponding by (11) to the morphism

p∗ (Φ) ⊗ q ∗ (tr−1 ) : p∗ E2 ⊗ q ∗ OP1 (2) → p∗ E1 [1] .

Now, by (14), SP1 (σC−1 (Φ)) is the element of Ext1C×P

b
∗ ∗ ∗
1 (p̂ S(E2 ) ⊗ q̂ OP1 (2), p̂ S(E1 )) corre-

sponding by (11) to the morphism

p̂∗ (S(Φ)) ⊗ q̂ ∗ (tr−1 ) : p̂∗ S(E2 ) ⊗ q̂ ∗ OP1 (2) → p̂∗ S(E1 )[1]

which, again by Remark 3.2, corresponds to σC−1

b (S(Φ)).

Remark 3.11. In order to ensure that the Fourier-Mukai transform gives rise to morphisms
between moduli spaces of triples one should check that the transform preserves families
of (IT) triples. This can be checked directly as in the usual case of families of sheaves,
alternatively one can use Remark 3.5 and note that the Fourier-Mukai transform is well-
behaved with respect to families and therefore induces morphisms between the moduli
spaces of SU(2)-equivariant sheaves. △

3.3 Preservation of stability for small α

+ +
Let Nαs+ (n1 , n2 , d1 , d2 ) be the moduli space of αm -stable triples with αm = αm + ǫ such that
m
+
ǫ > 0 and (αm , αm ] does not contain any critical value. (We assume that d1 /n1 ≥ d2 /n2 ,
since this is a necessary condition for the moduli space not to be empty.) One has the
following (Proposition 3.23 in [7]).

+
Proposition 3.12. If a triple T = (E1 , E2 , Φ) is αm -stable, E1 and E2 are semistable.
+
Conversely, if E1 and E2 are stable and Φ 6= 0 then T = (E1 , E2 , Φ) is αm -stable.

Proposition 3.13. If (n1 , d1) = 1, (n2 , d2 ) = 1 and d1 /n1 > d2 /n2 , the moduli space of
stable triples Nαs+ is isomorphic to a PN -fibration over MC (n1 , d1 ) × MC (n2 , d2 ), where
m

N = n2 d1 − n1 d2 − 1.

Proof. By Proposition 3.12, Nαs+ is the projectivization of a Picard sheaf on MC (n1 , d1 ) ×

m

MC (n2 , d2 ) (Corollary 6.2 in [8]), which in this case is a vector bundle with fibre H 0 (C, E2∗ ⊗
E1 ) over (E1 , E2 ), since H 1 (C, E2∨ ⊗ E1 ) ≃ H 0 (C, E1∨ ⊗ E2 )∗ = 0.
21
 Given an ITi triple T = (E1 , E2 , Φ) with transform Tb = (E
b1 , Eb2 , Φ)
b we denote by α̂m
the minimum value of the stability parameter α̂ with the type (n̂1 , n̂2 , dˆ1 , dˆ2 ) defined by Tb.
+ +
As above, α̂m is any real number such that the interval (α̂m , α̂m ] does not contain critical
values.

+
Theorem 3.14. Let T = (E1 , E2 , Φ) be a αm -stable triple such that (n1 , d1 ) = 1, (n2 , d2 ) =
1 and d1 d2 > 0 (this forces Φ 6= 0). Then the Fourier-Mukai transform Tb = (E b1 , E
b2 , Φ)
b
+
is α̂m -stable. The result also holds in the converse direction with the obvious modifications
on the hypotheses.

Proof. By Proposition 3.12 we have that E1 and E2 are both semistable. Moreover, E1
and E2 are stable due to the conditions on the rank and degree. Thus in the triple
Tb = (E
b1 , E
b2 , Φ)
b both bundles are stable. By Proposition 3.12 again we conclude that the
triple Tb is also αm
+
-stable. The proof of the converse is identical.

Corollary 3.15. Keeping the conditions stated in the previous Theorem and assuming
additionally that d1 /n1 > d2 /n2 , then the Fourier-Mukai transform induces an isomorphism

∼
→ Nα̂s+m .
S : Nαs+m −

In other words, the Fourier-Mukai transform induces an isomorphism between the PN -

fibrations described in Proposition 3.13.

3.4 Preservation of stability for large α

Recall that αL is the largest critical value in the interval (αm , αM ). If αL < α < αM the sta-
bility condition does not vary in this range, and we can then denote by Nαs− (n1 , n2 , d1 , d2 )
M
the moduli space of α-stable triples for any value α ∈ (αL , αM ).
The relationship between the stability of the triple and that of the involved bundles is
given by the following Proposition ([8, Propositions 7.5 and 7.6]).

Proposition 3.16. Let T = (E1 , E2 , Φ) be an α-semistable triple for some αL < α < αM ,
and let us suppose that n1 > n2 . Then T defines an extension of the form

Φ
0 → E2 −
→ E1 → F → 0 (15)
22
with F locally free, and E2 and F are semistable. Conversely, let T = (E1 , E2 , Φ) be a
triple defined by a non trivial extension of the form (15), with F locally free. If E2 and F
are stable then T is α-stable for αL < α < αM .

From this we have the following result (Theorem 7.7 in [8]).

Theorem 3.17. Let n1 > n2 , d1 /n1 > d2 /n2 , (n1 − n2 , d1 − d2 ) = 1 and (n2 , d2 ) = 1.
Then the moduli space Nαs− (n1 , n2 , d1, d2 ) is smooth of dimension n2 d1 − n1 d2 + 1 and it
M
is isomorphic to a PN -fibration over MC (n2 , d2 ) × MC (n1 − n2 , d1 − d2 ), whose fibre over
the point (E2 , F ) is given by PH 1 (C, E2 ⊗ F ∗ ), and N = n2 d1 − n1 d2 − 1.

Remark 3.18. The case n1 < n2 reduces to the situation in Theorem 3.17 by considering
the dual triple. △

We prove now that the Fourier-Mukai transform preserves stability for “large” values
of the parameter α.

Theorem 3.19. Let T = (E1 , E2 , Φ) be an α-stable triple such that (n1 − n2 , d1 − d2 ) = 1,

(n2 , d2) = 1, n1 6= n2 and αL < α < αM . Suppose also that d1 > 0, d2 > 0 and d1 − d2 > 0
(resp. d1 < 0, d2 < 0 and d1 −d2 < 0); then T is IT0 (resp. IT1 ) and the transformed triple
Tb = (E
b1 , E
b2 , Φ)
b is α̂-stable for α̂ ∈ (α̂L , α̂M ) where α̂L and α̂M are the values corresponding
to the transformed triple Tb.

Proof. We prove the IT0 case, the proof of the other case is entirely similar. Without
loss of generality we may assume n1 > n2 . By Proposition 3.16 the map Φ : E2 → E1 is
injective and the quotient sheaf F in 0 → E2 → E1 → F → 0 is locally free. Moreover, E2
and F are stable, and hence IT0 , from which it follows that E1 is IT0 . Transforming the
above sequence we get
b2 → E
0→E b1 → Fb → 0 .

Since the Fourier-Mukai transform preserves stability (Proposition 2.1) it follows that E b2
and Fb are stable. By Proposition 3.16 Tb = (E
b1 , E
b2 , Φ)
b is α̂-stable for α̂ ∈ (α̂L , α̂M ). The
proof of the converse is identical.

Under the same conditions of Theorem 3.19 we have the following.

23
Corollary 3.20. The Fourier-Mukai transform induces an isomorphism between the mod-
uli spaces of ITi stable triples:

Nαs− (n1 , n2 , d1 , d2 ) ≃ Nα̂s− ((−1)i d1 , (−1)i d2 , (−1)i+1 n1 , (−1)i+1 n2 ) .

M M

As a consequence, the Fourier-Mukai transform yields an isomorphism between the PN -

fibrations described in Theorem 3.17.

3.5 Applications to moduli spaces on C × P1

One notable application of the theory of triples is the construction of slope-stable bundles
on C × P1 with respect to the polarization ωα , with α > 0 (see Theorem 9.2 in [8]). It
seems quite natural to use the relative transform SP1 to further study the properties of
those bundles and to produce new examples of stable bundles. We give in this section a
result on the preservation of stability for a class of bundles on C × P1 which can not be
handled using the standard techniques based on choosing “suitable polarizations” as done
for example in [9] or [16], because the polarizations ωα are not suitable; the reason for this
being that there exist SU(2)-equivariant bundles which are ωα -stable and whose restriction
to a fibre, is never stable (here we are assuming Φ 6= 0). To see this, take E such that
Φ : E2 → E1 is not an isomorphism and note that the restriction of such a bundle to a fibre
Ct is given by an extension
0 → E1 → Et → E2 → 0 .
Since the associated triple is stable, Lemma 4.5 in [7] implies that Ext1 (E2 , E1 ) = 0
whenever Φ is not an isomorphism, therefore the previous extension is always split and
the restriction Et is not stable.
The following Proposition follows now immediately.

Proposition 3.21. Let T = (E1 , E2 , Φ) be an α-stable triple and let E be its associated
vector bundle on C × P1 . Then T is ITi if and only if E is ITi with respect to SP1 .

We can use this Proposition to prove the following result.

Theorem 3.22. Let T = (E1 , E2 , Φ) be an α-stable triple with E1 ≃ E2 and Φ 6= 0.

Assume that either rk(E1 ) = rk(E2 ) > 1 or deg E1 = deg E2 6= 0. Then the associ-
b is
ated SU(2)-equivariant bundle E on C × P1 is IT and the Fourier-Mukai transform E
polystable. Moreover, the triple Tb = (E
b1 , E
b2 , Φ)
b is α̂-stable for any α̂ > 0.
24
Proof. Thanks to Theorem 3.4 we have E ≃ (p∗ E1 ⊗ q ∗ OP1 (1)) ⊕ (p∗ E2 ⊗ q ∗ OP1 (1)) with
E1 ≃ E2 stable. The base change property for the Fourier-Mukai transform implies

SP1 (E) ≃ (p∗ S(E1 ) ⊗ q ∗ OP1 (1)) ⊕ (p∗ S(E2 ) ⊗ q ∗ OP1 (1))

Therefore, E is IT if and only if E1 ≃ E2 is IT with respect to S, and this follows from

Proposition 2.1 since the stability of E1 ≃ E2 implies that its degree is not zero unless the
rank is 1. The polystability of E b is a consequence of the above expression for E b and the
fact that S preserves stability, see Proposition 2.1.
On the other hand, let us recall that a triple (E1 , E2 , Φ) with E1 ≃ E2 is α-stable, for
any α > 0, if and only if Φ is an isomorphism and E1 ≃ E2 is stable [7, Lemma 4.6]. These
conditions are preserved by S, therefore Tb is α̂-stable for any α̂ > 0.

Collecting previous results, particularly Theorem 3.19, Theorem 3.14 and Theorem 3.4,
we have.

Theorem 3.23. Let T be an α-stable triple of type (n1 , n2 , d1, d2 ) with α and (ni , di )
satisfying one of the conditions

(i) (n1 − n2 , d1 − d2 ) = 1, (n2 , d2 ) = 1, n1 6= n2 and αL < α < αM . Suppose also that

di > 0 for i = 1, 2, d1 − d2 > 0 (resp. di < 0 i = 1, 2, d1 − d2 < 0) and αL < α < αM ,
+
(ii) (n1 , d1 ) = 1, (n2 , d2) = 1, d1 d2 > 0 and αm < α < αm ,

(i.e. one of the conditions in Theorems 3.19 or 3.14). Then, the corresponding SU(2)-
equivariant bundle E on C × P1 is IT. Moreover if E1 and E2 are not isomorphic, then
b is stable with respect to the polarization ωα̂ , where α̂ is the
the Fourier-Mukai transform E
corresponding parameter for the transformed triple according to Theorem 3.19 in case (i)
and to Theorem 3.14 in case (ii).

The relative Fourier-Mukai transform induces an isomorphism between the correspond-

ing moduli spaces of SU(2)-equivariant bundles as follows from the previous Theorem and
Remarks 3.5 and 3.11. Therefore we have.

Corollary 3.24. Let Nαs be a moduli space of α-stable triple satisfying one of the conditions
SU (2)
(i) or (ii). Let Mα be the corresponding moduli space of SU(2)-equivariant bundles on
1
C × P . Then the relative Fourier-Mukai transform gives an isomorphism
(2)∼ SU (2)
SP1 : MSU
α −
→ Mα̂ .
25
4 Nahm transforms for triples

4.1 Relative Nahm transform

In this section we modify the absolute Nahm transform to produce a relative version of it.
b =
For every elliptic curve C we consider the projections q : X = C × P1 → P1 , q̂ : X
b × P1 → P1 where C
C b is the dual elliptic curve. We endow the pull-back PP1 of the
b with the pull-back connection ∇P . For every point
Poincaré line bundle to X ×P1 X, P1
b × P1 we endow the Hermitian line bundle PP1 , x̂ ≡ PP1 |X
x̂ = (ξ, t) ∈ C → Xq̂(x̂) ,
q̂(x̂)
b of q over q̂(x̂) ∈ P1 , with the flat
obtained by restricting PP1 to the fiber Xq̂(x̂) ⊂ X ×P1 X
unitary connection ∇x̂ given by the restriction of ∇P . In this way X b parametrizes the
P1

gauge equivalence classes of Hermitian flat line bundles along the fibers of q : X → P1 .
Let us consider a Hermitian vector bundle E → X with a unitary connection ∇. We
denote by Et the restriction of E to the fibre Xt = q −1 (t), ∇t is the restriction of ∇ to
Et . On the vector bundle Eq̂(x̂) ⊗ PP1 , x̂ we have the connection ∇x̂ = ∇q̂(x̂) ⊗ 1 + 1 ⊗ ∇x̂ .
Therefore we have the family of coupled Dirac operators
√
Dx̂ = 2∂¯E∗ q̂(x̂) ⊗PP1 , x̂ : Ω0 (Xq̂(x̂) , Eq̂(x̂) ⊗ PP1 , x̂ ) → Ω0,1 (Xq̂(x̂) , Eq̂(x̂) ⊗ PP1 , x̂ ).

As in the absolute case we define the index Ind(D) of this family of Dirac operators D
parametrized by Cb × P1 . The relative Nahm transform maps a Hermitian vector bundle
with a unitary connection over C × P1 into a Hermitian vector bundle with a unitary
b × P1 .
connection over C

Definition 4.1. Let (E, ∇) be a pair formed by a Hermitian vector bundle E over C × P1
and a unitary connection ∇ on E. We say that (E, ∇) is an ITP1 (index Theorem) pair
relative to P1 if either Coker D = 0 or Ker D = 0. In the first case we say that (E, ∇) is
an ITP1 , 0 -pair, whereas in the second we call it an ITP1 , 1 pair. The transformed bundle of
b = ±Ind(D) → C
an ITP1 , i -pair is, according to the parity of i, the vector bundle E b × P1 .

Proceeding in the same way as in the absolute case we can endow the transformed vector
bundle of an ITP1 -pair with a Hermitian metric and a unitary connection in a natural way.
26
In doing this, since all the fibrations involved are trivial, the main difference one encounters
is that the parameter space of the family is enlarged from C b to C
b × P1 , but since Xq̂(x̂) ≃ C
the Dirac operators are still defined on vector bundles over the elliptic curve C. Therefore,
the theory parallels the one developed in the absolute setting.

b ∇)
Definition 4.2. Let (E, ∇) be an ITP1 -pair. We call (E, b the relative Nahm transform
of (E, ∇) and denote it by NP1 (E, ∇).

Let E → C × P1 be a holomorphic vector bundle endowed with a unitary connection

∇ compatible with the holomorphic structure. Since the spinc Dirac operator Dx̂ gets
identified with the Dolbeault-Dirac operator of Eq̂(x̂) ⊗ PP1 , x̂ , by Hodge theory and the
Dolbeault isomorphism we have

Ker Dx̂ ≃ H 0 (Xq̂(x̂) , Eq̂(x̂) ⊗ PP1 , x̂ ) (16)

Coker Dx̂ ≃ H 1 (Xq̂(x̂) , Eq̂(x̂) ⊗ PP1 , x̂ ). (17)

Let us suppose that E is ITi with respect to the relative Fourier-Mukai transform
described in Section 3.2. The isomorphisms (16) and (17) mean that (E, ∇) is an ITP1 , i -
pair. As we saw there, by [2, Theorem 2] or [12, Theorem 3.2.8] we have a natural C ∞
vector bundle isomorphism induced by Hodge theory

∼ b
φP1 : SPi 1 (E) −
→ E.

b is a holomorphic line bundle and the

Moreover, since the Poincaré bundle PP1 → C × C
connection ∇PP1 is compatible with the holomorphic structure, the same arguments that
in the absolute case led us to prove Theorem 2.12 give us now the following.

Theorem 4.3. Let F1 , F2 be Hermitian vector bundles over C × P1 endowed with unitary
connections ∇1 , ∇2 such that (F1 , ∇1 ), (F2 , ∇2 ) are ITP1 ,i -pairs with respect to the Nahm
transform. Then we have

1. The connections ∇ b 1, ∇
b 2 are compatible with the holomorphic structures of S i 1 (F1 ),
P
i b b
S 1 (F2 ), respectively. Thus, the curvature of the connections ∇1 , ∇2 is of type (1, 1).
P

2. For every holomorphic morphism Ψ : F1 → F2 we have an induced holomorphic mor-

27
 phism N (Φ) : Fb1 → Fb2 and a commutative diagram

φF 1
Fb1 / S(F1 ) (18)
NP1 (Ψ) SP1 (Ψ)
 φF 2 
Fb2 / S(F2 )

4.2 Relative Nahm transform for SU (2)-invariant Einstein-Her-

mitian connections

Let us suppose that E1 , E2 are complex Hermitian vector bundles over C and let us choose
an SU(2)-invariant metric on H ⊗2 . We put on the bundle E = p∗ E1 ⊕ (p∗ E2 ⊗ q ∗ H ⊗2 ) the
Hermitian metric which is determined in a natural way by the Hermitian metrics of E1 ,
E2 and H ⊗2 .
By Proposition 3.5 in [14] there is a one to one correspondence between the SU(2)-
invariant unitary connections on E and the triples T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) formed by
unitary connections ∇1 , ∇2 on E1 , E2 , respectively, and a C ∞ vector bundle morphism
Φ : E2 → E1 . Moreover, this correspondence also holds at the level of SU(2)-invariant
holomorphic structures on E. Before discussing it we introduce the following.

Definition 4.4. We call a triple T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) integrable if Φ : E2 → E1 is

holomorphic with respect to the holomorphic structures determined by the connections ∇1
and ∇2 .

Proposition 3.9 in [14] gives us a one to one correspondence between SU(2)-invariant

holomorphic structures on E, considered as integrable SU(2)-invariant connections (i.e.
connections with curvature of type (1, 1)), and integrable triples T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ).
This is precisely the content of Proposition 3.1 which gives us a bijective correspon-
dence between the SU(2)-invariant holomorphic structures on E and holomorphic triples
T = (E1 = (E1 , ∂¯∇1 ), E2 = (E2 , ∂¯∇2 ), Φ).
Let us denote by ∇T the SU(2)-invariant integrable connection on E determined by
an integrable triple T . If we express its curvature with respect to the splitting E =
28
p∗ E1 ⊕ (p∗ E2 ⊗ q ∗ H 2 ) we have

!
∇T p∗ F ∇ 1 − β ∧ β ∗ ∂β
F = (19)
−∂β¯ ∗ ′
p∗ F ∇2 ⊗ 1 + 1 ⊗ q ∗ F ∇ − β ∗ ∧ β,
′
where F ∇i is the curvature of the connection ∇i , F ∇ is the curvature of the unique SU(2)-
invariant unitary connection on H ⊗2 , β = p∗ Φ ⊗ q ∗ η, with η an SU(2)-invariant section of
H ⊗−2 and ∂¯ is the Cauchy-Riemann operator determined by the connections ∇1 , ∇2 and
∇′ , for further details see [14, 7].
We want to study the relative Nahm transform of the SU(2)-equivariant bundles
(E, ∇T ) associated to integrable triples.
The following is straightforward.

Proposition 4.5. Let T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) be an integrable triple on C and let

(E, ∇T ) be its associated bundle with connection over C × P1 . If both (E1 , ∇1 ), (E2 , ∇2 )
are ITi -pairs then (E, ∇T ) is an ITP1 , i pair.

Given an integrable triple T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) such that (E1 , ∇1 ) and (E2 , ∇2 )
are ITi -pairs we can form the triple Tb = ((E
b1 , ∇
b 1 ), (E
b2 , ∇
b 2 ), Φ)
b obtained by means of the
absolute Nahm transform. Here we have denoted by Φ b the Nahm transform N (Φ). By the
sake of brevity the same notation is used hereafter. On the other hand, if (E, ∇T ) is the
vector bundle with connection over C × P1 associated to the triple T , we can apply to it
the relative Nahm transform to obtain NP1 (E, ∇T ). Taking into account the compatibility
between the Fourier-Mukai and Nahm transforms, Theorems 2.12 and 4.3 and Proposition
3.10 we have.
b × P1 associated to the triple
Proposition 4.6. NP1 (E, ∇T ) is the vector bundle on C
Tb = ((E
b1 , ∇
b 1 ), (E
b2 , ∇
b 2 ), Φ).
b

Definition 4.7. Let T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) be an integrable triple on C. We say that

it satisfies the τ -coupled vortex equations if

iΛF ∇1 + ΦΦ∗ = 2πτ IdE1

iΛF ∇2 − Φ∗ Φ = 2πτ ′ IdE2 ,

Note that in order to have solutions τ , τ ′ must fulfill the following equation

n1 τ + n2 τ ′ = d1 + d2 , (20)
29
with ni = rank(Ei ) and di = deg(Ei ).
The following Proposition was proved in [13] (see also [7]).

Proposition 4.8. Let T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) be an integrable triple and let ∇T be the
corresponding connection on E. Let τ and τ ′ be related by (20) and let us suppose that
(n1 + n2 )τ − d1 − d2
α= > 0.
n2
Then T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) satisfies the τ -coupled vortex equations if and only if
∇T is a Einstein-Hermitian connection on E → C × P1 with respect to the Kähler form
ωα = α2 p∗ ωC + q ∗ ωP1 , where ωP1 is the Fubini-Study Kähler form normalized to volume one
and ωC is a Kähler form of unit volume.

Proposition 4.9. Let T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) be an integrable triple on C which sat-

isfies the τ -coupled vortex equations and let X = C × P1 . Then:

(i) If the Hermitian endomorphisms 2πτ IdE1 −ΦΦ∗ and 2πτ ′ IdE2 +Φ∗ Φ are non-negative
and there exist x1 , x2 ∈ C such that 2πτ IdE1 −ΦΦ∗ (x1 ) > 0, 2πτ ′ IdE2 +Φ∗ Φ(x2 ) > 0,
then (E, ∇T ) is an ITP1 , 0 -pair and (E1 , ∇1 ), (E2 , ∇2 ) are IT0 pairs.

(ii) If the Hermitian endomorphisms 2πτ IdE1 −ΦΦ∗ and 2πτ ′ IdE2 +Φ∗ Φ are non-positive
and there exist x1 , x2 ∈ C such that 2πτ IdE1 −ΦΦ∗ (x1 ) < 0, πτ ′ IdE2 +Φ∗ Φ(x2 ) < 0,
then (E, ∇T ) is an ITP1 , 1 -pair and (E1 , ∇1 ), (E2 , ∇2 ) are IT1 pairs.

Proof. For every x̂ = (ξ, t) the restriction of E = p∗ E1 ⊕ (p∗ E2 ⊗ q ∗ H ⊗2) to Xq̂(x̂) ≃ C is

isomorphic to E1 ⊕ E2 as C ∞ bundles. Now (19) implies that the curvature of ∇q̂(x̂) with
respect to the splitting Eq̂(x̂) ≃ E1 ⊕ E2 is
!
F ∇1 0
F ∇q̂(x̂) = .
0 F ∇2
The claim now follows from Theorem 2.6.

4.3 Covariantly constant triples

Definition 4.10. Let T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) be an integrable triple on C. We will

say that T is covariantly constant if ΦΦ∗ is covariantly constant with respect to ∇1 and
Φ∗ Φ is covariantly constant with respect to ∇2 .
30
Remark 4.11. Denote by ∇ the connection naturally induced on Hom(E2 , E1 ) by ∇1 and
∇2 . If Φ is covariantly constant with respect to ∇ then it is easy to check that T is
covariantly constant. Moreover, Φ is covariantly constant with respect to ∇ if and only if
Φ : E2 → E1 is an anti-holomorphic map.

Proposition 4.12. Let T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) be a covariantly constant integrable

triple on C. Then we have holomorphic orthogonal decompositions

E1 ≃ Ker Φ∗ ⊕ E1′
E2 ≃ Ker Φ ⊕ E2′

which are compatible with the connections, and Φ induces an holomorphic isomorphism
Φ : E2′ → E1′ .

Proof. Since ΦΦ∗ and Φ∗ Φ are covariantly constant vector bundle endomorphisms, they
are holomorphic and their eigenvalues are constant. Moreover, ΦΦ∗ , Φ∗ Φ are positive Her-
mitian endomorphisms whose spectrum may differ only at 0; therefore we have orthogonal
decompositions

E1 = Ker Φ∗ ⊕ E1 (λ1 ) ⊕ · · · ⊕ E1 (λk )

E2 = Ker Φ ⊕ E2 (λ1 ) ⊕ · · · ⊕ E2 (λk ),

where E1 (λi ), E2 (λi ) are the eigenbundles with eigenvalue λi 6= 0 with respect to the
holomorphic endomorphisms ΦΦ∗ and Φ∗ Φ, respectively. Since these endomorphisms are
covariantly constant, the subbundles E1 (λi ), E2 (λi ) are preserved by the connections ∇1 ,
∇2 , respectively. Moreover, for every λi we have an isomorphism
∼
Φ : E2 (λi ) −
→ E1 (λi )

Therefore if we denote E1′ = E1 (λ1 ) ⊕ · · · ⊕ E1 (λk ), E2′ = E2 (λ1 ) ⊕ · · · ⊕ E2 (λk ), we have

an isomorphism
∼
→ E1′
Φ : E2′ −

as required.

With the same notations as above we have the following

31
Proposition 4.13. Let T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) be a covariantly constant integrable
triple on C. Then T satisfies the τ -coupled equations if and only if

1. ∇1 induces a constant central curvature connection on Ker Φ∗ with factor 2πτ , unless
Ker Φ∗ = 0, and a constant central curvature connection on E1′ with factor π(τ + τ ′ )
unless E1′ = 0.

2. ∇2 induces a constant central curvature connection on Ker Φ with factor 2πτ ′ unless
Ker Φ = 0 and a constant central curvature connection on E2′ with factor π(τ + τ ′ )
unless E2′ = 0.

Proof. Since T is covariantly constant we have the decompositions

E1 = Ker Φ∗ ⊕ E1 (λ1 ) ⊕ · · · ⊕ E1 (λk )

E2 = Ker Φ ⊕ E2 (λ1 ) ⊕ · · · ⊕ E2 (λk ),

provided by Proposition 4.12. Moreover, since T satisfies the τ -coupled equations we have

iΛF ∇1 = 2πτ IdE1 −ΦΦ∗

iΛF ∇2 = 2πτ ′ IdE2 +Φ∗ Φ.

Therefore we have

iΛF ∇1 | Ker Φ∗ = 2πτ IdKer Φ∗ iΛF ∇1 |E1(λi ) = (2πτ − λi ) IdE1 (λi ) (21)
iΛF ∇2 | Ker Φ = 2πτ ′ IdKer Φ iΛF ∇2 |E2(λi ) = (2πτ ′ + λi ) IdE2 (λi ) . (22)

This implies that Ker Φ∗ , E1 (λi ), Ker Φ, E2 (λi ) are bundles with constant central cur-
vature connection with slopes
λi
µ(Ker Φ∗ ) = τ µ(E1 (λi )) = τ −
2π
λi
µ(Ker Φ) = τ ′ µ(E2 (λi )) = τ ′ + .
2π
But since E1 (λi ) is isomorphic to E2 (λi ) we must have µ(E1 (λi )) = µ(E2 (λi )), that is
λi = π(τ − τ ′ ). Therefore in the above decompositions there is only one eigenvalue and if
we substitute λi in 21-22 we get the required values for the factors of the constant central
curvature connections. The converse statement is just a simple checking.
32
Corollary 4.14. Let T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) be a covariantly constant integrable triple
on C which satisfies the τ -coupled equations. Then T is (τ − τ ′ )-polystable. Moreover, T
decomposes as a sum of (τ − τ ′ )-polystable triples.

T = (Ker Φ∗ , 0, 0) ⊕ (E1′ , E2′ , Φ) ⊕ (0, Ker Φ, 0).

Proof. The Hitchin-Kobayashi correspondence for triples, [7, Theorem 5.1], establishes
an equivalence between triples that satisfy the τ -coupled vortex equations and (τ − τ ′ )-
polystable triples. Therefore, the Corollary follows at once.
However, in the present case it is possible to give a direct proof. Since T satisfies the
τ -coupled vortex equations, Proposition 4.13 implies that Ker Φ∗ , E1′ ≃ E2′ and Ker Φ are
polystable bundles with slopes
1
µ(Ker Φ∗ ) = τ, µ(E1′ ) = µ(E2′ ) = (τ + τ ′ ), µ(Ker Φ) = τ ′ .
2
Therefore we have

µα (Ker Φ∗ , 0, 0) = µ(Ker Φ∗ ) = τ
α
µα (E1′ , E2′ , Φ) = µ(E2′ ) = µ(E1′ ) + = τ
2
µα (0, Ker Φ, 0) = µ(Ker Φ) + α = τ ,

where α = (τ −τ ′ ). Since E2′ carries a constant central curvature, there exists an orthogonal
decomposition
(1) (m)
E2′ = E2 ⊕ · · · ⊕ E2

compatible with the connection and such that every factor carries an irreducible constant
central curvature connection. Since Φ∗ Φ = λ IdE2 it follows that we have an orthogonal
decomposition
(1) (m)
E1′ = Φ(E2 ) ⊕ · · · ⊕ Φ(E2 )
(i) (i)
Thus, the triple (E1′ , E2′ , Φ|E2′ ) splits into the direct sum of subtriples (E2 , Φ(E2 ), Φ|E (i) )
2
(i)
with E2 stable and Φ|E (i) an isomorphism. By [7, Proposition 3.21] this implies that
2
(E1′ , E2′ , Φ|E2′ ) is α-polystable. Therefore, T is α-polystable (see [7, Definition 3.15]).

Remark 4.15. If T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) is a covariantly constant integrable triple on

C which is α-stable with E1 6= 0 and E2 6= 0, then the previous Corollary implies that Φ
has to be an isomorphism. △
33
 As a consequence of Proposition 4.9 we immediately obtain.

Lemma 4.16. Let T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) be a covariantly constant integrable triple

on C which satisfies the τ -coupled equations.

(i) If τ > 0 and τ ′ > 0 then (E, ∇T ) is an ITP1 , 0 -pair and (E1 , ∇1 ), (E2 , ∇2 ) are IT0
pairs.

(ii) If τ < 0 and τ ′ < 0 then (E, ∇T ) is an ITP1 , 1 -pair and (E1 , ∇1 ), (E2 , ∇2 ) are IT1
pairs.

Theorem 4.17. Let T = ((E1 , ∇1 ), (E2 , ∇2 ), Φ) be a covariantly constant integrable triple

on C which satisfies the τ -coupled equations and let (E, ∇T ) be its associated bundle with
connection over C × P1 .

1. If τ > 0 and τ ′ > 0 then the Nahm transform Tb = ((E b1 , ∇

b 1 ), (E
b2 , ∇
b 2 ), Φ)
b is a
covariantly constant integrable triple. Moreover, Tb satisfies the τ̂ -coupled equations,
for some value of τ̂ , if and only if τ = τ ′ .

2. If τ < 0 and τ ′ < 0 then the Nahm transform Tb = ((E b1 , ∇

b 1 ), (E
b2 , ∇
b 2 ), Φ)
b is a
covariantly constant integrable triple. Moreover, Tb satisfies the τ̂ -coupled equations,
for some value of τ̂ , if and only if τ = τ ′ .

Proof. Proposition 4.12 gives us a decomposition

E1 ≃ Ker Φ∗ ⊕ E1′
E2 ≃ Ker Φ ⊕ E2′

Since T satisfies the τ -couple vortex equations, Proposition 4.13 implies that (Ker Φ∗ , ∇1 ),
(E1′ , ∇1 ) ≃ (E2′ , ∇2 ) and (Ker Φ, ∇2 ) are bundles with constant central curvature with
slopes µ(Ker Φ∗ ) = τ , µ(E1′ ) = µ(E2′ ) = 12 (τ + τ ′ ), µ(Ker Φ) = τ ′ . Now if we apply the
Nahm transform and denote (Φ) b ∗ by Φ
b ∗ , Theorem 2.11 implies that (Ker Φ b ∗ = Ker
\ b 1 ),
Φ∗ , ∇
b′ , ∇
(E b 1 ) ≃ (E
b′ , ∇
b 2 ) and (Ker Φ
b = Ker
\ b 2 ) are bundles with constant central curvature
Φ, ∇
1 2
and we get a decomposition
b1 ≃ Ker Φ
E c′
b∗ ⊕ E
1
b2 ≃ Ker Φ
E c′
b ⊕E2

34
 b ∗ Φ)
The conditions (Φ∗ Φ)|E2′ = λ IdE2′ , (ΦΦ∗ )|E1′ = λ IdE1′ with λ 6= 0 imply (Φ b b ′ = λ Id b ′ ,
|E2 E2
b b ∗ b
(ΦΦ )|Eb ′ = λ IdEb ′ . Let us prove the first equality in the IT0 case. Given s, t ∈ E2,ξ =
1 1
∇2,ξ 0
Ker ∂ ⊂ Ω (E2,ξ ) one has

b ∗ Φ(s),
hΦ b b
tiEb2,ξ = hΦ(s), b
Φ(t)i b2,ξ
E

Taking into account the definition of the Hermitian metric on Eb2,ξ given in (3) of Section
b given in Theorem 2.12 we get
2.2 and the definition of Φ
Z Z
b b
hΦ(s), Φ(t)iEb2,ξ = hΦ(s), Φ(t)iE2 ω = hΦ∗ Φ(s), tiE2 ω
Cξ Cξ

b ′ one has
Therefore, if s, t ∈ E2,ξ

b ∗ Φ(s),
hΦ b tiEb2,ξ = λ hs, tiEb2,ξ

which proves our claim. The second equality follows in the same way. The proofs in the
IT1 case are entirely similar.
This proves that ((E b1 , ∇
b 1 ), (E
b2 , ∇
b 2 ), Φ)
b is a covariantly constant integrable triple. More-
over, the slopes of these bundles are µ(Ker Φ b ∗ ) = − 1 , µ(E
b′ ) = µ(E
b ′ ) = − 2 ′ , µ(Ker Φ)
b =
τ 1 2 τ +τ
− τ1′ . An easy computation shows now that Tb fulfills the conditions of Proposition 4.13 in
order to have a solution of the τ̂ -coupled vortex equations, for some value of τ̂ , if and only
if τ = τ ′ .

As a consequence of the preceding Theorem and the Hitchin-Kobayashi correspondence

for triples (Theorem 5.1 in [7]), which establishes an equivalence between holomorphic
triples which satisfy the τ -coupled equations and α-polystable triples, we have.

Corollary 4.18. Polystability is not preserved, in general, under the Fourier-Mukai and
Nahm transform.

Proof. It is enough to take any stable bundles F1 , F2 , F such that µ(F ) = 12 (µ(F1 )+µ(F2 ))
and µ(F1 ) > µ(F2 ), which are known to exist since the moduli spaces of stable bundles with
fixed coprime rank and degree over an elliptic curve C are isomorphic to C and thus they
are not empty (see [27]). Now define the triple T = (F1 , 0, 0) ⊕ (F, F, IdF ) ⊕ (0, F2 , 0) and
endow F1 , F2 , F with connections of constant curvature compatible with their holomorphic
structures according to Donaldson Theorem [11]. Now, Proposition 4.13 implies that T
35
is (τ − τ ′ )-polystable since, by construction, it satisfies the τ -coupled equations, with
τ = µ(F1 ) and τ ′ = µ(F2 ).
If we take µ(F1 ) 6= µ(F2 ), Theorem 4.17 implies that the transformed triple Tb does not
satisfy the τ̂ -coupled equations for any value of τ̂ . By the Hitchin-Kobayashi correspon-
dence for triples [7, Theorem 5.1], this implies that Tb is not polystable.

The preservation of stability remains as an open question. Notice that in the case
of stable triples (E1 , E2 , Φ) with E1 6= 0 and E2 6= 0, the condition of being covariantly
constant implies that Φ is an isomorphism (Remark 4.15). Now stability is preserved in
the conditions of Theorem 3.22.
Acknowledgement. The authors are deeply grateful to the anonymous referee for
his/her remarks which have helped to significantly improve the paper.

References
[1] M.F. Atiyah, Vector bundles over an elliptic curve, Proc. London Math. Soc. 7 (1957),
414-452.

[2] C. Bartocci, U. Bruzzo, D. Hernández Ruipérez, Fourier-Mukai transform and index

theory, Manuscripta Math. 85 (1994), 141–163.

[3] C. Bartocci, U. Bruzzo, D. Hernández Ruipérez, A Fourier-Mukai transform for

stable bundles on K3 surfaces, J. Reine Angew. Math. 486 (1997), 1–16.

[4] C. Bartocci, U. Bruzzo, D. Hernández Ruipérez, A hyper-Kähler Fourier transform,

Differential Geom. Appl. 8 (1998),239–249.

[5] J.M. Bismut, The Atiyah-Singer index for families of Dirac operators: Two heat
equation proofs, Invent. Math. 83, (1986), 91–151.

[6] P. J. Braam, P. van Baal, Nahm’s transformation for instantons, Comm. Math. Phys.
122 (1989), 267–280.

[7] S. Bradlow, O. Garcı́a-Prada, Stable triples, equivariant bundles and dimensional

reduction, Math. Ann. 304 (1996), 225-252.
36
 [8] S. Bradlow, O. Garcı́a-Prada, P. Gothen, Moduli spaces of holomorphic triples over
compact Riemann surfaces, Math. Ann. 328 (2004), 299-351.

[9] T. Bridgeland, Fourier-Mukai transforms for elliptic surfaces, J. reine angew. Math.
498 (1998), 115–133.

[10] M. Dekker, Moduli spaces of stable sheaves on abelian surfaces, Ph.D. Thesis, Uni-
versity of Amsterdam (1997).

[11] S.K. Donaldson, A new proof of a theorem of Narasimhan and Seshadri. J. Differential
Geom. 18 (1983), no. 2, 269–277.

[12] S.K. Donaldson, P.B. Kronheimer, The geometry of four manifolds, Oxford Mathe-
matical Monographs, Clarendon Press, Oxford (1991).

[13] O. Garcı́a-Prada, The geometry of the vortex equation, D.Phil. Thesis, Oxford (1991).

[14] O. Garcı́a-Prada, Dimensional reduction of stable bundles, vortices and stable pairs,
International Journal of mathematics Vol 5, 1 (1994) 1-52.

[15] R. Hartshorne, Residues and duality, Lecture Notes in Mathematics 20, Springer-
Verlag, Berlin (1966).

[16] D. Hernández Ruipérez, and J.M. Muñoz Porras, Stable sheaves on elliptic fibrations,
Journal of Geometry and Physics, 43 (2002) 163-183.

[17] N.J. Hitchin, On the construction of monopoles, Comm. Math. Phys. 89 (1983), no.
2, 145–190.

[18] M. Jardim, Construction of doubly-periodic instantons, Comm. Math. Phys. 216

(2001), 1–15.

[19] M. Jardim, Nahm transform and spectral curves for doubly-periodic instantons,
Comm. Math. Phys. 225 (2002) 639–668.

[20] S. Kobayashi, Differential geometry of complex vector bundles, Publications of the

Mathematical Society of Japan, 15. Kanô Memorial Lectures, 5. Princeton University
Press, Princeton, NJ; Iwanami Shoten, Tokyo, (1987).
37
[21] A. Maciocia, Gieseker stability and the Fourier-Mukai transform for abelian surfaces.
Quart. J. Math. Oxford Ser. (2) 47 (1996), no. 185, 87–100.

[22] S. Mukai, Duality between D(X) and D(X̂) with its application to Picard sheaves,
Nagoya Math. J. 81 (1981), 153–175.

[23] S. Mukai, Fourier functor and its application to the moduli of bundles on an abelian
variety, Adv. Pure Math. 10 (1987), 515-550.

[24] W. Nahm, The construction of all self-dual multimonopoles by the ADHM method.
Monopoles in quantum field theory (Trieste, 1981), 87–94, World Sci. Publishing,
Singapore, 1982.

[25] M.S. Narasimhan and C.S. Seshadri, Stable and unitary bundles on a compact Rie-
mann surface, Ann. Math. 82 (1965), 540-564.

[26] K. O’Grady, The weight-two Hodge structure of moduli spaces of sheaves on a K3

surface, J. Algebraic Geom. 6 (1997), no. 4, 599–644.

[27] L. W. Tu, Semistable bundles over an elliptic curve, Adv. in Math.98 (1993) 1-26.

[28] K. Yoshioka, Moduli spaces of stable sheaves on abelian surfaces Math. Ann. 321
(2001), no. 4, 817–884.

38