Treball final de grau GRAU DE MATEMATIQUES` Facultat de Matem`atiques Universitat de Barcelona
TWISTOR THEORY AND GRAVITATIONAL INSTANTONS
Autor: Alejandro Fern´andezPiqu´e
Director: Dr. Ignasi Mundet Riera Realitzat a: Departament d’Algebra` i Geometria
Barcelona, 27 de Juny de 2016 Abstract
The aim of this work is to understand and introduce a self-contained account of the construction of certain kind of gravitational instantons made by Hitchin in his work “Polygons and Gravitons”. Gravitational instantons are solutions of the Einstein equations which have a complete, non-singular and positive definite metric. Hitchin use twistor theory to construct these solutions. We present beforehand some concepts of self-duality, differential operators and connections to understand twistor theory. Afterwards, the construction of the aforementioned instantons is explained in full detail.
Acknowledgements
I would like to thank my advisor Ignasi Mundet Riera for all the answers and detailed explanations he has given me. I also would like to emphasize his patience which has been tested after being asked every week. I also thank my family and friends for being always by my side. Finally, I would like to thank Ainhoa for being so supportive and understanding.
i Contents
1 Preface 1
2 Riemannian and Spin Geometry 2 2.1 Spinors and p-forms ...... 2 2.2 Self-duality ...... 6
3 Twistor non-linear graviton 8 3.1 Frobenius and Newlander-Nirenberg theorems ...... 8 3.2 First order differential operators and jet bundles ...... 9 3.3 Integrability of the Twistor space ...... 13 3.4 Properties of the twistor space of a self-dual manifold ...... 20
4 Construction of Gravitational Instantons 22 4.1 Gravitational Instantons ...... 22 4.2 Hitchin’s Construction ...... 22
5 Conclusions 33
A Appendix. Basic Material 34 A.1 Lie Groups ...... 34 A.2 Clifford Algebras and Spin Groups ...... 36 A.3 Representation Theory ...... 39 A.4 Fibre Bundles ...... 40 A.5 Connections in Fibre Bundles ...... 42 A.6 Complex Geometry ...... 45
ii 1 Preface
The project
Yang-Mills instantons are non-perturbative solutions of the Yang-Mills equations which helped resolve some of the problems that appeared during the 1970s. Due to their relevance in the path integral quantization, similar solutions for Euclidean quantum gravity were studied. They were called gravitational instantons. The Yang-Mills instantons that had self-dual curvature were classified and stud- ied using algebraic geometry and also using the ideas introduced by Penrose and Ward. This methodology let the interpretation of a 4-dimensional Minkowski space in terms of a 3-dimensional complex space. This methodology is called twistor theory. A similar approach can be taken with gravitational instantons. If the curvature derived from the metric is self-dual, it is posible to use the “non-linear graviton” ideas of Penrose to convert the problem of finding such metrics as one of complex geometry. Hitchin used this technique to construct solutions of Einstein equations called ”gravitational multi-instantons”, which were introduced previously by Gib- bons and Hawking [11]. The aim of this “Treball de Fi de Grau” is to give a self-contained explanation of the non-linear graviton introduced by Penrose and to construct the gravitational multi-instanton metric as Hitchin did. The main reason for taking this amazing journey is to give a more mathematical perspective to the previous “Treball de Fi de Grau” of the Physics Degree, which focused also on gravitational instantons. In order to explain all the concepts of this construction in full detail and in a reader-friendly manner, the text is divided into three sections:
1. The first section gives an introduction of Riemann geometry. It gives an elementary view on Clifford algebra, spinors and their representations. It also introduces the notion of self-duality for differential forms, for Riemann manifolds and for connections.
2. The aim of the second section is to understand two very important proposi- tions that will let us define the twistor space of a manifold. For this purpose, an introduction to first order differential equations and jet bundles is given.
3. Finally, on the third section we explain the construction made by Hitchin with the tools of the previous sections. We construct the metric making some assumptions that are proved subsequently.
In the appendix there are some concepts that will be necessary through these sec- tions in order to understand the explanations. It is divided into topics and contain definitions and statements that will be needed through the construction.
1 2 Riemannian and Spin Geometry
2.1 Spinors and p-forms
Let M be an oriented Riemannian manifold of an even dimension 2l, i.e. a smooth manifold with a smooth inner product defined on the tangent space of each m M. Let Λp denote the bundle of exterior p-forms1 . It is posible to define the Hodge∈ star operator :Λp Λ2l−p as: ∗ → α β = α, β ω ∧ ∗ h i where ω denotes the volume form2, α, β Λp and , the induced inner product on p-forms by the metric on the manifold.∈ This innerh producti is determined by the condition: v1 vp, w1 wp = det ( vi, wj ) h ∧ · · · ∧ ∧ · · · ∧ i h i If we consider the case l = 2, then 2 = ( 1)p(2l−p)s = ( 1)p(4−p)s, where s is the signature of the inner product. For an∗ oriented− Riemannian− manifold, the signature p 2 2 is s = 1 and the vector space Λ for p = 2, Λ , splits into the direct sum of Λ+ and 2 Λ−, the subspaces of the self-dual and anti-self-dual 2-forms respectively. Now we will introduce another notion involving the Hodge star that will let us later define the desired manifolds properly. The curvature of a manifold of dimension n 4 can be decomposed into three irreducible terms. Furthermore, if n = 4 one of the≥ terms decomposes into two terms . Taking the curvature tensor as a 2 2 2 2 2 linear map from Λ Λ and recalling the decomposition of 2-forms, Λ = Λ+ Λ−, we can view the curvature→ tensor as the following matrix: ⊕ s W+ + Id Z 12 (2.1) t s Z W− + 12 Id
where s is the scalar curvature, Z is the traceless Ricci tensor and W+ and W− together give the Weyl curvature W = W+ + W− . The terms on the upper left- hand side and lower right-hand side are self-adjoint [4]. The 2-forms can be identified, using the metric, with the Lie Algebra of SO(4), so(4), i.e. with skew adjoint transformations of Λ1. This identification can be derived by associating to a pair of vectors v, w Λ1 the skew-symmetric endomor- phism v w defined by: ∈ ∧ (v w)(x) = v, x w w, x v ∧ h i − h i 1 A covariant p-tensor on a real vector space V (in out case is T ∗M) is said to be alternating if any permutation of the arguments causes its value to be multiplied by the sign of the permutation. Alternating covariant p-tensor are called p-forms. The wedge product ( ) acts as follows: a b = a b b a. ∧ ∧ ⊗ − ⊗ p 2In local coordinates it is described by ω = g dx dx . | | 1 ∧ · · · ∧ 2l 2 The resulting map Λ2 so(4) is a vector space isomorphism. The notion of Lie bracket can be applied to→ 2-forms making the previous identification an isomorphism 2 2 2 of Lie algebras. Moreover, the decomposition described before, Λ = Λ+ Λ−, corresponds to the isomorphism of Lie Algebras so(4) = so(3) so(3). We⊕ can 2 ⊕ see thereby that Λ± are 3-dimensional real vector spaces which are isomorphic as vector spaces to so(3), the vector space of skew-symmetric matrices of dimension 3 . Moreover, we can make this isomorphism a Lie algebra isomorphism. The 3- dimensional real vector space have an inner product derived from the metric on M. We can define then the vectorial product as: (a, b) a b = a b, where is the Hodge dual. This also gives the Lie bracket to this→ space× becoming∗ ∧ thereby∗ a Lie algebra. We have obtained then the identification Λ2 3 so(3) due to the + ∼= R ∼= vectorial product. There is also an important identification in 4-dimensional manifolds. The uni- versal cover of SO(4), Spin(4), is not a simple group3 but decomposes as
Spin(4) = SU(2) SU(2) × In order to understand this equality, R4 must be identified by the quaternions H (see Definition A.1.2). Taking the elements p, q of SU(2) = Sp(1) (see Definition A.1.3) as unit quaternions, the map x pxq−1 is an orthogonal transformation of the quaternions with unit determinant→ which has as kernel the element ( 1, 1). Therefore, SU(2) SU(2) is a doble covering of SO(4) and it implies Spin− (4)− = SU(2) SU(2). × × Due to the previous identification, the representations of Spin(4) can be ex- pressed using the representations of SU(2). The fundamental representation D1/2 is SU(2) acting on C2 and all other irreducible representations are symmetric powers k + Dk/2 = S D1/2, with k Z . The tensor product of two representations decom- poses using Clebsch-Gordan∈ formula (A.3.3) :
Dk/2 Dl/2 = Dk+l/2 Dk+l−2/2 D|k−l|/2 (2.2) ⊗ ⊕ ⊕ · · · ⊕ So the representation ring4 of Spin(4) is based on the two fundamental repre- sentations D± , where we use to distinguish the representation of each of the 1/2 ± SU(2), Spin(4) = SU(2)+ SU(2)−. The representation D+ D− has dimen- × k/2 ⊗ l/2 sion (k + 1)(l + 1) and factors through SO(4) if and only if k + l is even. These are orthogonal irreducible representations of Spin(4) and all others are symplectic 56 .
It is posible to define, at least locally, the complex vector bundles V+ and V− which are associated to the principal spin bundle through irreducible representations + of Spin(4). The vector bundle V+ is associated with the representation D C 1/2 ⊗ 3A simple group is a nontrivial group whose only normal subgroups are the trivial group and the group itself. 4It is the ring formed from all the finite-dimensional linear representations of the group. 5If V is an irreducible representation of G and W is an irreducible representation of H, then V W is an irreducible representation of G H, see [6, page 82]. ⊗6A symplectic representation of a Lie Group× G into V is a representation that preserves the symplectic form ω of V , i.e ω(ga, gb) = ω(a, b) for a, b V . ∈
3 − of Spin(4) = SU(2) SU(2) and V− with the representation C D . The total × ⊗ 1/2 spin bundle is defined as V = V+ V−, it is a Z2-graded vector bundle. Spinors are defined, consequently, as sections⊕ of these fibre bundles.
The Z2-graded vector spaces are also called superspaces. If the superspace is the direct sum of two complex vector spaces with Hermitian structures, it is called a Her- mitian superspace. A superalgebra A is an algebra whose underlying vector space i j i+j is a superspace and whose product respects the Z2-grading, i.e. A A A . The Clifford algebra7 C`(Λ1, g) is a superalgebra, C`(Λ1, g) = C`0(Λ1,· g) ⊂C`1(Λ1, g). The algebra of endomorphisms End(E) of a superspace E is a superalgebra,⊕ such that: End+(E) = Hom(E+,E+) Hom(E−,E−) End−(E) = Hom(E+,E−) ⊕ Hom(E−,E+) ⊕ If E is a Hermitian superspace, we say that u End−(E) is odd self-adjoint if it has the form ∈ 0 u− u = u+ 0 where u+ : E+ E− and u− is the adjoint of u+. → A very important proposition concerning modules8 of the Clifford algebra of the tangent space of an 2l-dimensional manifold is the next one: Proposition 2.1. If E is a 2l-dimensional oriented Euclidean vector space , then + − there is a unique Z2-graded Clifford algebra module S = S S , called the spinor bundle , such that ⊕ C`(V ) C = End(S) ⊗ ∼ Since Spin(2l) C`(E), it follows that both S+ and S− are representations of Spin(2l). They⊂ are called half-spinor representations.
See [3, page 109] for the proof. Therefore, the previous representations of Spin(4) can be also understood as restrictions to Spin(4) of the representations of the Clifford bundle. ∗ p At the same time, the complexified Clifford algebra is isomorphic to Λc = pΛc p p ⊕ as a graded vector bundle, where Λc = Λ C [16, page 95][10, page 13]. Hence, we have the following vector bundles isomorphisms:⊗
∗ ∗ 1 1 Λ (T M = Λ ) = C`c(Λ , g) = End(V = V+ V−) c ∼ ∼ ⊕ Therefore, it is posible to associate the p-forms to endomorphisms of the spinors. They act on spinors in the following way: Λ1 Hom(V ,V ) Hom(V ,V ) Λ3 c ∼= + − ∼= − + ∼= c 7An introduction to Clifford algebras and spinors can be found at the Appendix A.2. 8 + We say that E = E E− is a Clifford module if there is a bundle map c : T ∗M End(E) such that: ⊕ → c(α )c(α ) + c(α )c(α ) = 2(α )(α ) • 1 2 2 1 − 1 2 + c(α) swaps the bundle E and E− •
4 1 Moreover, the real bundle of Λc can be identified with Hom(V+,V−) Hom(V−,V+). 2 ⊕ On the other hand, Λ+c End(V+) are the traceless endomorphisms of V+ and the ⊂ 9 2 real bundle are those which are also skew hermitian . In the case of Λ−c it works on the opposite way. As we have seen before, Λ2 so(3) which implies that Λ2 so(3, ). Recalling + ∼= +c ∼= C that V+ is the vector bundle obtained by the fundamental representation of SU(2), 2 we know that there is a volume form ν in V+ which is invariant by the action of ∧2 SU(2). We can represent each element of S (V+), i.e. the symmetric tensor product 2 2 of V+, as a polynomial ax + bxy + cy . Therefore, we can use the discriminant of 2 b2−ac the polynomial to give a Euclidean inner product to S (V+) such that ν C. In this way, we can use this inner product to define the Hodge dual and to define∈ the 2 notion of Lie Algebra by [a, b] = a b for a, b S (V+). This algebra has complex ∗ ∧ ∈ 2 2 dimension 3 and is isomorphic to so(3, C). Hence, Λ+c = S V+. This also makes ∗ 2 ∼ ∗ sense taking into account that V+ = V+,Λ+c End(V+) = V+ V+ = V+ V+, 2 ∼ ⊂ ∼ ⊗ ∼ ⊗ S V+ V+ V+ and have the same dimension. ⊂ ⊗ An important remark that will be used in the next section derives from the previous discussion. The bundle of self-dual Weyl tensors is identified with the 10 2 bundle of self-adjoint traceless endomorphisms of Λ+. The bundle of all endo- 2 2 morphisms of this vector space is Λ+ Λ+ and using the Clebsch-Gordan formula 2 2 2 2 4 ⊗ 2 0 (2.2), Λ+ Λ+ = S V+ S V+ = S V+ S V+ S V+ are obtained. As previously 2⊗ ∼ ⊗ ∼ ⊕ ⊕ stated, S V+ is the bundle of skew hermitian transformations and hence it can not 0 be identified with the Weyl term. S V+ is a one-dimensional bundle of scalar trans- 4 formations which are not clearly traceless. Thus, S V+ is the bundle of self-dual − 4 Weyl tensors. Similarly, W Γ(S V−). ∈ 1 1 Another important point of Λ Λ = Hom(V+,V−) is their relation with the ⊂ c ∼ almost complex structures on M. Fixing a non-zero spinor φ (V+)m at m gives a real isomorphism Λ1 (V ) defined by the Clifford multiplication∈ α α φ and m ∼= − m in turn this identification gives a complex structure on the tangent space→ of· M at m. This complex structure is compatible with the metric and the orientation of M. Multiplying φ by a scalar λ C∗ does not modify this complex structure, hence ∈ P (V+)m parametrizes the set of compatible complex structures. The same can be done using V−, but in this case, the orientation is the inverse. Finally, it is important to realize that the fundamental representation of SU(2), i.e. D1/2, has a symplectic structure, because SU(2) SL(2, C) and a unitary structure because SU(2) U(2). Therefore, ≤ ≤ ¯ V+ = V+ = V+ ∼ ∗ ∼ Hence, the composition is a quaternionic structure:
j : V+ V+ → where j(λv) = λj¯ (v) and j2 = 1. − 9 It is minus its conjugate transpose, S = S†. 10A self-adjoint operator A on a vector space− V with inner product , is that which satisfies for all x, y V : Ax, y = x, Ay . h i ∈ h i h i 5 2.2 Self-duality
The Hodge star operator is conformally invariant11 on l-forms, where 2l is the dimension of the manifold. If we multiply the metric by a factor λ, then the result of the product of two vectors is multiplied by λ and for 1-forms by λ−1. Consequently, the result is multiplied by λ−l on l-forms. The volume form is multiplied by λl so (α, β)ω = α β remains invariant. ∧ ∗ Recalling the decomposition of the curvature in (2.1) , we can define a self-dual manifold as follows:
Definition 2.1. An oriented Riemannian 4-manifold is self-dual if its Weyl tensor W = W+, i.e if W− = 0.
The Weyl tensor is also conformally invariant, therefore this is a property of the conformal structure. A very important notion in differential geometry is the connection or the covari- ant derivative. A brief introduction to connections on fibre bundles can be found at the Appendix (A.5). The notion of equivalence between connections is called gauge equivalence or gauge invariant, i.e the invariance of the connection under a gauge transformation:
Definition 2.2. A gauge field transformation on a principal G-bundle P is a dif- feomorphism f : P P such that (1) f(gp) = gf(p) g G, p P , (2) f preserves each fibre, i.e acts trivially→ on the base space M. ∈ ∈
As we did for manifolds, we can also characterize the self-duality of a connection:
Definition 2.3. On a 4-manifold M, a connection is said to be self-dual if its curvature Ω is in A2 (g)12 (i.e Ω = Ω) and anti-self-dual if Ω A2 (g) (Ω = Ω). + ∗ ∈ − −∗ Finally, the next proposition will clarify the properties of the 4-dimensional man- ifolds that will be used and the kind of connections of the fibre bundles on the manifolds.
Proposition 2.2. Let M be a 4-manifold with an Einstein metric13 . Then the induced connections on the bundle of self-dual spinors V+ and the bundle of self-dual 2 2-forms Λ+ are self-dual. The induced connections on the corresponding anti-self- dual bundles are anti-self-dual. Conversely, if the induced connections on V+ and 2 Λ+ are self-dual, the metric is an Einstein metric.
2 Proof. The vector bundle of self-dual forms Λ+ defined over M has an associated principal bundle called the frame bundle which has SO(3) as a Lie group, the group 2 of rotations of Λ+. If we take a metric on M we can induce a connection on this
11It is invariant under conformal transformations, see Definition A.4.13. 12 k Vk 2 2 A (E) denotes the bundle of sections in Γ(E T ∗M). In this case, A = Γ(g Λ+). 13A metric is Einstein if its Ricci tensor is proporcional⊗ to the metric. However, in this⊗ context we refer to Einstein metrics as those whose Ricci tensor also vanishes.
6 principal bundle from the Levi-Civita connection of M. The adjoint bundle is the associated vector bundle14 obtained due to the adjoint representation15 of the Lie 2 group. The lie algebra of SO(3), so(3), can be identified with Λ+ as we have seen before. Therefore, the adjoint bundle of this principal bundle is the vector bundle 2 2 Λ+. The curvature then is a section of A (g), which is the part of the Riemann 2 2 curvature tensor which lives in Λ+ Λ . Looking at the matrix representation of the curvature tensor (2.1), we see that⊗ the curvature on the vector bundle is the first row: s Ω = W + Id + Z + 12 2 2 Since Z is the only term in A−(Λ+), the connection on the vector bundle is self-dual if and only if Z = 0, i.e. if the manifold is an Einstein manifold. The self-duality property of the connection doesn’t change between associated bundles, hence V+ is also self-dual if M has an Einstein metric and is a spin manifold.
14see Definition A.4.7 . 15see Definition A.1.7 .
7 3 Twistor non-linear graviton
Most of the work done by Hitchin at [12] relies on two statements of “Self-Duality in Four-Dimensional Riemannian Geometry” [1]. The first one is the Proposition 3.3 which gives two necessary and sufficient conditions for V (D¯) T ∗E∗ (a vector subspace derived from a differential operator D¯ that acts on the vector⊂ space E) to be involutive. Afterwards, Theorem 3.3 makes use of the aforementioned proposition to define an integrable almost complex structure on P (V−), a bundle defined over M, which is the vanishing of W−. These two statements will let us define the twistor space. Some tools of differen- tial geometry are needed in order to prove them and understand their implications.
3.1 Frobenius and Newlander-Nirenberg theorems
First, it is necessary to introduce the concept of distribution and some properties that can be applied to it. Due to length restrictions, most of the statements will be given without proof. Our aim is to recall some concepts that will be used later on. Definition 3.1. Let c be an integer, 1 c d.A c-dimensional distribution D on ≤ ≤ a d-dimensional manifold M is a choice of a c-dimensional subspace Dm of TmM for each m M. D is smooth if for each m in M there is a neighborhood U of ∈ ∞ m and there are c vector field X1,...Xc of class C on U which span D at each point of U. A vector field X on M is said to belong to the distribution D(X D) ∈ if Xm Dm for each m M. A smooth distribution D is called involutive if [X,Y ] ∈D whenever X and∈ Y are smooth vectors fields lying in D. ∈ Definition 3.2. A submanifold (N, ψ) of M where ψ is an embedding is an integral manifold of a distribution D on M if
dψ(Nn) = D(ψ(n)) for each n N ∈ Definition 3.3. A smooth distribution D is said to be integrable if each point of M is contained in an integral manifold of D Definition 3.4. A smooth distribution D TM is completely integrable if there exist a flat chart for D in a neighborhood of⊂ each point of M. Proposition 3.1. Every integrable smooth distribution is involutive
See [22, page 42] for the proof.
In order to understand the proposition (3.3), it is necessary to define the distri- bution in terms of differential forms. Lema 3.1. Suppose M is a smooth n-manifold and D TM a distribution of rank k. Then D is smooth if and only if each point m of M⊂ has a neighborhood U on which there are smooth 1-forms ω1, . . . , ωn−k such that for each q U: ∈ 1 n−k Dq = kerω q ker ω q | ∩ · · · ∩ | 8 See [17, page 493] for the proof.
On the next proposition we will find equivalent conditions for a distribution to be involutive in terms of differential forms. Proposition 3.2. Let D be a smooth distribution of rank k on a smooth n-manifold M, and let ω1, . . . , ωn−k be a smooth defining forms for D on an open U M. The following statements are equivalent: ⊂
1. D is involutive on U
2. dω1, . . . , dωn−k annihilate D
3. There exists smooth 1-forms αi : i, j = 1, . . . , n k such that j − n−k i X j i dw = w αj j=1 ∧
See [17, page 495] for the proof.
Finally, let us introduce two very important theorems which, with the help of the implications of the aforementioned statements, will enable the construction of a complex manifold. Theorem 3.1. (Frobenius). Every involutive distribution is completely integrable.
See [22, page 43] for the proof. Remark 3.1. Therefore, completely integrable integrable involutive ⇔ ⇔ Theorem 3.2. (Newlander and Nirenberg). Let (M,J) be a 2l-dimensional almost complex manifold. If J is integrable, the manifold M is a complex manifold with the almost complex structure J.
See [19] for the proof.
3.2 First order differential operators and jet bundles
There are two ways to understand the proposition 3.3 and one of them is in terms of jet bundles. Let us begin by defining the notion of a differential operator: Definition 3.5. A differential operator of order k on M is a linear map P : Γ(E) Γ(F ), where E and F are smooth complex vector bundles over M, with the following→ property. Each point of M has a neighborhood U with local coordinates (x1..., xn) p q and local trivializations: E U U C and F U U C , in which P can be written in the form: | → × | → × X ∂|α| P = Aα(x) ∂xα |α|≤k
9 where each Aα(x) is a q p-matrix of smooth complex-valued functions and where Aα = 0 for some α with×α = k. A real differential operator of order k is defined 6 | | similarly with C replaced by R.
Another notion related to a differential operator is the symbol map: ∂|α| Definition 3.6. The Symbol map σ of a differential operator P = P Aα(x) |α|≤k ∂xα of order k, is described by: X σ(P ) = Aα(x)ψα |α|=k where ψ = (dx1, . . . , dxr). It can be defined intrinsically: taking a covector v ∗ ∞ ∈ Tx M, a section se of E such se(x) = e Ex and a function f C (M) such f(x) = 0 and df = v for x M , then: ∈ ∈ ∈ k D(f se)(x) = σ(D)(x)(v) Fx · ∈ Remark 3.2. From now on, only first-order differential operators will be considered, i.e k = 1.
Now, let us define a new type of fibre bundle, the jet bundle, that will be very important to understand the first proposition. First of all, we will define an equiv- alence relation and then we will used it to define the fibre bundle.
Definition 3.7. Let us consider a fibre bundle π : E M, where M has dimension n. Taking m M, denote Γ(E) all the sections of E→that contain m in the domain. Let us define∈ an equivalence relation in Γ(E). Two sections σ and η have the same k-jet at m if |I| α |I| α ∂ σ ∂ η I = I , 0 I k. ∂x m ∂x m ≤ | | ≤ where I is any ordered n-tuple of integers I = (I(1),I(2),...I(n)), such that
n m I(i) X ∂|I| Y ∂ I = I(i) , := . ∂xI ∂xi | | i i=1
k The k-jet with representative σ is denoted by jσ or jk(σ). Remark 3.3. By using the previous definition we can define a k-th jet manifold J k(E) as: J k(E) := jrσ : p M, σ Γ(E) p ∈ ∈ and it defines fibre bundle with the projections:
( r ( r πr : J (E) M πr,0 : J (E) E → , → jrσ p jrσ σ(p) p 7→ p 7→
10 Let us define now the subspace V (D¯) T ∗(E∗ 0) mentioned at the beginning of the section. Let E be a real vector bundle⊂ on M\ . Section s Γ(E) defines by duality a function sv C∞(E∗) on the dual bundle such that: ∈ ∈ v s (m) = s(m), m h i where m is a point of M defined by x1, . . . , xn . Let us see these operators in local { } ∗ terms: taking the basis (e1, . . . , ek) for E and (1, . . . , k) for E , a parametrization of E and E∗ can be done locally by:
k X (f1, . . . , fk, x1, . . . , xn) fi(m)ei(m) → i
k X (λ1, . . . , λk, x1, . . . , xn) λi(m)ei(m) → i v P v The action of s for any s = i fiei and the 1-form obtained by derivation of s has the form: k v X s (λ1, . . . , λk, x1, . . . , xn) = λifi(m) i k k v X X ds = dλifi + λidfi i i The homomorphism from sections of E, Γ(M,E) to covectors of the vector space E∗, i.e to Γ(E∗,T ∗E∗) also factors through the 1-jet bundle. Taking p : E∗ M as the projection, one obtains the following diagram. →
duality ∗ d ∗ ∗ ∗ Γ(M,E) ∼ Γ(E , R) / Γ(E ,T E ) = 6 V
∗ p ∗ Γ(M,J1(E)) / Γ(E ,J1(E))
∗ ∗ ∗ ∗ The homomorphism V : Γ(E ,J1(E)) Γ(E ,T E ) is characterized by the prop- erty → ∗ v V (p j1(s)) = ds and is surjective off the zero section of E∗. This can be interpreted as follows: ∗ the zero section m = 0 of E is determined by λi(m) = 0, i, therefore for every ∗ v ∀ p j1(s) m=0, the image through V is the 1-forms ds , which at this point m = 0 | v Pk ∗ ∗ ∗ ∗ has the form ds = i dλifi. However, the subspace p (T M) T E is generated by dxi, for i = 1, . . . , n. Taking dλ1, . . . dλn, dx1, . . . , dxn as a⊂ basis for T ∗E∗, we ∗ ∗ ∗ { } see that p (T M) V (p (J1(E))) = 0. ∩ ∗ 1 ∗ The homomorphism V restricted to p (E Λ ) p (J1(E)) gives: ⊗ ⊂ ∗ ∗ ∗ V ((em αm) ) = em, m p αm (T E ) (3.1) ⊗ m h i ∈ m
11 ∗ 1 ∗ Let us prove this statement. The fibre of the sub-bundle p (E Λ ) p (J1(E)) 1 ⊗ ⊂ (we can do this identification J1(E) = E E Λ ) at m is formed by those 1-jets ∼ ⊕ ⊗ whose representatives vanish at m, i.e. those which vanish at m. In local terms, those sections are determined by the condition fi(m) = 0, i. The image by V of v Pk Pk P∀n dfi r this sub-bundle is then 1-forms ds = i λidfi = i λi r dxr dx . Contracting k s P this 1-form to a vector X = ∂x + i ai∂λi ,
k X dfi dsv(X) = λ i dxs i
1 Pn Pk fj i An element of E Λ will be of the form ( ( i ej) dx ) which applied to V ⊗ i j dx ⊗ ∗ Pn Pk ∂fj ∗ l Pn Pk ∂fj l using (3.1) gives em, m p αm = l j λj ∂xl p (dx ) = l j λj ∂xl dx . Applying this 1-form to Xh: i k X dfj e , p∗α (X) = λ m m m j dxs h i j which proves the statement. Therefore, a linear first-order differential operator D¯ is defined as a homomor- phism from J(E) to F , whose kernel R is taken to be a vector bundle. Then the vector bundle V (D¯) defined in the proposition is just V (p∗R). We can also define it ∗ ∗ as the sub-bundle of the cotangent space of E 0 such that at each point m E 0, v \ ∈ \ this vector bundle is formed by 1-forms (ds )m derived from sections s Γ(M,E) ¯ ∈ which D(s)m = 0. Let us introduce an important example that will be of main use for the proof of the first proposition. Exemple 3.1. Let us take D¯ = . Taking as ξ any section of E, the covariant derivative16 applied to this section∇ gives as follows:
k k k k ¯ X X X X Dξ = D(ξαeα) = eα dξα + eβ ωαβξα ⊗ ⊗ α=1 α=1 α=1 β=1 where the property (fs) = df s + f s ∇ ⊗ ∇ is used on the second equality. The covariant derivative of the section is zero iff for any α: β dξα + wβαξ = 0 (3.2) v P P The image through V of this sections are ds = dλiξi + λidξi. Using the condition (3.2) gives as follows:
v X X X X X X X X ds = dλiξi λiwjiξj = dλiξi λjwijξi = (dλi λjwij)ξi i − i j i − i j i − j P Therefore, θi = dλi wijλj span V ( ). − j ∇ 16 e = P w e . ∇ i j ij ⊗ j 12 3.3 Integrability of the Twistor space
The two statements introduced at the beginning of the prior section will be proved in this section. Let us take D¯ as a general differential operator of the form σ 1 ∇ where σ is the symbol of the operator. Let S1 = R E Λ be the kernel of the 2 1 ∩ ⊗ 17 symbol and S2 E Λ the image of S1 Λ under exterior multiplication . Let us recall the first⊂ proposition:⊗ ⊗
¯ ∗ ∗ Proposition 3.3. V (D) T (E 0) is involutive iff (1) D1Γ(S1) Γ(S2), (2) ⊂ 0 \ 2 ⊂ 1 ΩΓ(E) Γ(S2), where Ω: A (E) A (E) is the curvature of and D1 : A (E) A2(E) the⊂ extended covariant derivative.→ (Note that the first condition∇ is a “torsion”→ condition on the connection - the vanishing of certain components of the connection matrix - and the second condition is a condition on the curvature)
Proof. As we have seen in Proposition 3.2, the subspace V (D¯) will be involutive ∗ ¯ P and therefore involutive if for any section s Γ(E ,V (D)), then ds = si αi for ∗ ¯ ∗ ∗ ∗ ∈ ∧ vi Γ(E ,V (D)) and αi Γ(E ,T (E 0)). The kernel of the differential operator is the∈ direct sum of the kernel∈ of the covariant\ derivative, E, and the kernel of the ¯ ∗ symbol, S1. Therefore, V (D) = V (p (E S1)). This subspace is spanned by: ⊕ X θi = dλi wijλj , 1 i k − j ≤ ≤
v X σ = sijkλjdxk , 1 i m i ≤ ≤ j,k P where σi = j,k sijkej dxk is taken as a local basis of S1 and we have used the ⊗ v equation (3.1) to compute σi . Taking the tuple θ1, . . . , θk, dx1, . . . , dxn as the ∗ ∗ 2 18 { 1 } basis of T (E 0), the subspace V2 ΛE∗ (the image of V ΛE∗ under exterior \ ⊂ v ⊗ v multiplication) is spanned by θi θj, θi dxj, σ dxj. The term σ θj is not ∧ ∧ i ∧ i ∧ added because this term is also generated by θi dxj. S2 is generated by σi dxj, v ∧ ∗ 1 ∧ thus the forms σ dxj span V (S2) V2, where V (S2) is V (p S1) ΛM . Taking the exterior derivative∧ of the generators⊂ of V (D¯): ⊗ X X dθi = λjdwij dλj i j = − j − j ∧ X X X X = λjdwij wjk wijλk θj wij − − ∧ − ∧ j j k j X X = λkΩik θj wij − − ∧ k j where the last equality follows from: ! X X X X a b λk dwik + wij wjk = λkΩik = λkΩik,abdx dx (3.3) ∧ ∧ k j k k
17We use the relation a b b a = a b. 18 ⊗ − ⊗ ∧ We express explicitly with this notation that the domain of the sections is E∗.
13 P Since the element θj wij is a linear combination of elements θi dxj, dθi V2 P j ∧ ∧ ∈ if and only if k λkΩik V2. If this 2-form is in V2 it implies that is a linear ∈ v combination of θi θj, θi dxj and σ dxj, i.e. exist αu,v, α¯u,¯ v¯ andα ˜u,˜ v˜ such as: ∧ ∧ i ∧ k k k n n n n n k X X X X X X X X X αuvθu θv+ α¯u¯v¯θu¯ dxv¯+ α˜u˜v˜θu˜ θv˜ = λcΩic,abdxa dxb ∧ ∧ ∧ ∧ u v u¯ v¯ u˜ v˜ a b c
Thus, evaluating (∂λr , ∂λs ) on both sides of the equality implies:
αrs = αsr
Pk Pk Taking any r, s 1, . . . k and recalling that θu θv = θv θu gives αuvθu ∈ ∧ − ∧ u v ∧ θv = 0. Now, evaluating (∂λr , ∂xs ) on both sides again implies:
α¯rs = 0
Taking all r 1, . . . k and s 1, . . . , n we finally arrive to the point where; if (3.3) ∈ ∈ v is in V2, then it is a linear combination of only σ dxj. Therefore: i ∧ X X dθi Γ(V2) λkΩik Γ(V2) λkΩik Γ(V (S2)) ΩΓ(E) Γ(S2) ∈ ⇔ ∈ ⇔ ∈ ⇔ ⊂ k k On the other hand,
v X X dσ = sijkdλj dxk + λjdsijk dxk = i ∧ ∧ j,k j,k X X X = sijkdθj dxk + sijkλmwjm dxk + λjdsijk dxk = ∧ ∧ ∧ (3.4) j,k j,k,m j,k X v = sijkdθj dxk + (D1σi) ∧ j,k where it is used that X X D1σi = dsijk ej dxk + sijkD1(ej dxk) = ⊗ ⊗ ⊗ jk jk X X = dsijk ej dxk + sijk ej dxk = ⊗ ⊗ ∇ ⊗ jk jk ! X X X = dsijk ej dxk + sijk wjm em dxk = ⊗ ⊗ ⊗ ⊗ jk jk m ! X X X = dsijk ej dxk + simk wmj ej dxk ⊗ ⊗ ⊗ ⊗ jk mk j ! X X = dsijk + simkwmj ej dxk ⊗ ⊗ jk m
14 which combined with (3.1) gives: ! v X X (D1σi) = λj dsijk + simkwmj dxk ∧ jk m X X = sijkλmwjm dxk + λjdsijk dxk ∧ ∧ j,k,m j,k X X = sijkλmwjm,adxa dxk + λjdsijk dxk ∧ ∧ j,k,m,a j,k
v v Similarly to the previous case, (D1σi) could only be spanned by σi dxj. Therefore, the following result is obtained: ∧
v v v dσ Γ(V2) (D1σi) Γ(V2) (D1σi) Γ(V (S2)) D1Γ(S1) Γ(S2) i ∈ ⇔ ∈ ⇔ ∈ ⇔ ⊂ Remarks 3.1. 1. In the case D¯ = the symbol operator is the identity and ∇ therefore S1 = 0. This implies that for V ( ) to be involutive, the curvature must be zero. ∇
2. The proposition still holds for a complex vector space E if the complex struc- ture commutes with the covariant derivative and the differential operator. In this case, V (D¯) is a sub-bundle of the complexified cotangent bundle T ∗(E∗ 0). c \ ∗ ¯ 3. In the complex case, if the distribution V Tc is such that V V = 0 and ¯ ∗ ⊂ ∩ V + V = Tc , an almost complex structure J can be defined: J acts on the elements of V by multiplying by i and by i for the elements of V¯ . It can be also defined on the opposite way. If it− is involutive, then because of the Theorem 3.2, the structure is integrable.
Let us use this proposition to prove the next theorem:
Theorem 3.3. Let M be an oriented 4-manifold . Then a conformal structure on M defines in a natural way an almost complex structure on P (V−), which is integrable iff W− = 0, i.e iff M is self-dual. (Note that the spin representations are well defined projective representations of SO(4), so we need not assume that M is a spin manifold)
Proof. Let us consider the spin bundle V− and a metric within the conformal struc- ture. The Dirac operator is defined by:
∇ 1 σ D : Γ(V−) Γ(V− Λ ) Γ(V+) −→ ⊗ −→ where σ is the Clifford multiplication. On the other hand, the twistor operator D¯ is defined by: ¯ ∇ 1 σ¯ ⊥ D : Γ(V−) Γ(V− Λ ) Γ(V ) −→ ⊗ −→ +
15 ∗ 1 whereσ ¯ is the orthogonal projection 1 σσ onto the kernel of σ in Γ(V− Λ ). This projectionσ ¯ =: (V , Λ1) Ker(σ−) is described locally by [2]: ⊗ ± → 1 X σ¯(t ψ) = t ψ + e e t ψ n i i ⊗ ⊗ i ⊗ · · where n is the dimension of the manifold. Therefore, ! ¯ X X Dψ =σ ¯ ( ψ) =σ ¯ ( iψ dxi) = (¯σ( iψ dxi)) = ∇ i ∇ ⊗ i ∇ ⊗ ! X 1 X = ψ dx + e e ψ e = i i 4 j j i i i ∇ ⊗ j ⊗ · ∇ · 1 X = ψ + ei Dψ ei ∇ 4 · ⊗ P ¯ where it is used that Dψ = i ei ei ψ. Now let us focus on the properties of D. · ∇ 1 The kernel of the symbol, i.e. S1 are those elements of V+ embedded in V− Λ by the Clifford multiplication: ⊗ ! X X 1 X X σ¯( e ψ e ) = e ψ e + e e e e ψ = j j j j 4 i i j j j · ⊗ j · ⊗ i ⊗ j · · ! X 1 X X X X = e ψ e + e e 1 ψ = e ψ e e ψ e = 0 j j 4 i i j j i i j · ⊗ i ⊗ j − · j · ⊗ − i · ⊗ ∗ Using (3.1) at a point φ V , the elements of V (S1) are ∈ − X X V ( ej ψ ej) = ej ψ, φ ej j · ⊗ j h · i for each ψ Γ(V+). This is the subspace of holomorphic 1-forms of the complexified ∈ 1,0 1 ¯ ∗ ∗ cotangent space of M, i.e Λ Λc . Thus, V (D) = V (p S1 p V−) and is a 4- ⊂ ∗ ∗ ¯ ⊕ ¯ dimensional complex sub-bundle of Tc (V− 0) such that V (D) V (D) = 0. As it has been seen above, this defines an almost\ complex structure,∩ that applying the previous proposition, will be integrable. Let us see if this sub-bundle verifies the conditions of the proposition.
P P 1. Taking φ = ei ψ ei S1 where ψ V+, let us see that (φ) = ei · ⊗ ∈ ∈ ∇ · ψ ei: Doing the covariant derivative of φ we obtain: ∇ ⊗ X X X φ = ei ψ ei + ei ψ ei + ei ψ ei ∇ ∇ · ⊗ · ∇ ⊗ · ⊗ ∇ We need to prove that the sum of the first item and the third item equals 0. X X ei ψ ei + ei ψ ei = 0 (3.5) ∇ · ⊗ · ⊗ ∇
16 This comes from the compatibility of the connection with the metric, i.e.: ! X X g = 0 ei ei = 0 ( ei ei + ei ei) = 0 (3.6) ∇ ⇔ ∇ i ⊗ ⇔ i ∇ ⊗ ⊗ ∇
Now, the Clifford multiplication by an element ψ V+ is a lineal operation: ∈ ∗ ∗ ∗ ⊗V+ ∗ ∗ ∗ · ∗ ∗ T M T M T M T M V+ T M T M V− T M T M ⊗ ⊗ −−→ ⊗ ⊗ ⊗ −→ ⊗ ⊗ Therefore, multiplying φ to the compatibility condition (3.6) of the metric and realizing that this is a linear operator we prove (3.5) and again: X (φ) = ei ψ ei ∇ · ∇ ⊗ Hence, X D1φ = A( φ) = ei jψ ej ei Γ(S2) ∇ · ∇ ⊗ ∧ ∈ This is also seen by direct computation: ! X X X D1φ =D1 ei ψ ei = D1(ei ψ ei) = ( (ei ψ) ei + ei ψ dei) = i · ⊗ i · ⊗ i ∇ · ∧ · ⊗ X = ( j(ei ψ) ej ei + ei ψ dei) i,j ∇ · ⊗ ∧ · ⊗
Now, using that dei = 0, i and that the connection is torsion free, i.e. iej = ∀ ∇ jei, we have: ∇ X D1φ = (ei jψ ej ei) i,j · ∇ ⊗ ∧
1 2. The elements of Γ(S2) are generated by the image of the map V+ Λ 2 P ⊗ → V− Λ defined by ψ α ei ψ ei α. If this map is injective, then ⊗ 1 2 ⊗ → · ⊗ ∧ V+ Λ V− Λ . Let us see then that the kernel of the map is zero. If P ⊗ ⊂ ⊗ P ψj ej is in the kernel, then ei ψj ei ej = 0. This implies eiψj = ejψi for i =⊗j following that: · ⊗ ∧ 6
eiψj = ejψi eieiψj = eiejψi ψj = eiejψi = ejeiψi ⇒ ⇒ − − ⇒
ejψj = ejejeiψi = ejψj = eiψi ⇒ − This implies that eiψi = 0 which at the same time is equivalent to ψi = 0. Let us prove this by contraposition. If ψi = 0, then ψi = 0 eieiψi = 0 6 − 6 ⇒ 6 and thus eiψi = 0. 6 The map is injective. Therefore
1 2 V+ Λ V− Λ ⊗ ⊂ ⊗ Decomposing these two bundles into irreducible components:
1 2 0 2 V+ Λ = V+ V+ V− = (S V+ S V+) V− = (S V+ V−) V− ⊗ ∼ ⊗ ⊗ ∼ ⊕ ⊗ ∼ ⊗ ⊕ 17 2 2 2 2 2 2 3 V− Λ = V− (Λ Λ ) = (S V+ V−) (S V− V−) = (S V+ V−) V− S V− ⊗ ∼ ⊗ +⊕ − ∼ ⊗ ⊕ ⊗ ∼ ⊗ ⊕ ⊕ it can bee seen that ΩΓ(V−) Γ(S2) if and only if the composition ⊂ Ω 2 Re 3 V− V− Λ S V− −−→ ⊗ −−→ 3 is zero, i.e. the restriction Re to S V−. That would imply that the image 1 is in V+ Λ and therefore, an element of Γ(S2). The homomorphisms from 3⊗ 3 ∗ 3 V− S V− correspond to elements of S V− V− and thus also of V− S V−, → ∗ 4 ⊗ 3 ⊗ because V− = V−. We have that S V− V− S V− and we need this ∼ ⊂ 4⊗ homomorphism to vanish, therefore, the part S V− of the curvature must vanish. This part corresponds tp W− as we have seen in the first section. Therefore, the condition requieres W− = 0. ∗ This complex structure is defined on V− 0. Due to the invariance under scalar x \ ∗ multiplication λ C , this complex structure can be also defined on P (V−) = ∗ ∈ ¯ P (V−). The kernel of the twistor operator D is conformally invariant, hence the complex structure in determined by the conformal class of the metric.
Remarks 3.2. 1. This fibre bundle π : P (V−) M with the complex structure is defined as the twistor space of M, also denoted→ by . It represents the Z fibre bundle where each element Jx is a complex structure on TxM which gives the opposite orientation. Therefore,∈ Z a section of defines an almost complex structure on M. If the complex structure is involutive,Z then it is a 3-dimensional complex manifold. The twistor space is also identified with π : S√ (Λ2 ) M, where S√ is the subspace of Λ2 withZ norm equal to √2. 2 − → 2 − 2. The construction of the complex structure can be explained more geometrically. Fixing a metric in the conformal structure splits the tangent bundle of P (V−) into a vertical and horizontal subspace:
∗ TP (V−) = VP (V−) π TM ⊕
Recalling that V− is a complex vector bundle, it is easy to see that on the vertical subspace VP (V−), the complex structure is defined by the fibres. On the other hand, the complex structure on the horizontal part is defined as follows: Each point φ P (V−)m over m M defines a complex structure ∈ ∈∗ on TmM because of the real isomorphism T M V+ given by α α φ. → → · This complex structure defines the opposite orientation on M. Calling Jm ∗ this complex structure, the complex structure Jφ on π TM is:
∗ JφX = π (Jm)X = Jm(π∗X)
∗ for X π TM φ. ∈ |
3. A similar construction can be done using V+, but in this case, the complex structures in M preserve the orientation. For the almost complex structure on P (V+) to be involutive, it is necessary in this case W+ = 0 [23].
18 4. The fibres of π : P (V−) M are complex submanifolds. Each fibre is a → rational curve, i.e. a copy of the complex projective line CP1 and has as a normal bundle N = (1) (1) (see Definition A.6.7). Let us see this: we ∗ O ⊕ O take fibre `m = (V−)m with λ1, λ2 as linear coordinates. The conormal bundle ∗ v v N at this fibre is spanned by σ1 and σ2 , where:
4 v X σα = ei ψα, φ ei i h · i
∗ ∗ ∗ and ψα (V+)m, φ (V−)m. Both are holomorphic sections of N on (V−)x since, as∈ seen in (3.4)∈ , do not have dλ¯ terms. Thus, they trivialize N ∗ on ∗ ∗ (V−)m. Since they are linear in φ, on P (V−) they trivialize (1)N and ∗ O therefore N = (1) (1) or equivalently N = (V )m (1). O ⊕ O + ⊕ O 0 5. Holomorphic sections H (`m, (1)) are parametrized by (V−)m, hence: O 0 ∗ 1 H (`m,N) = (V V−)m = Hom(V+,V−)m = (Λ )m ∼ + ⊗ ∼ ∼ c as seen in the first section.
∗ 0 6. A decomposable element φ ψ (V V−)m defines a section φlψ of H (`, N) ⊗ ∈ + ⊗ where lψ is the linear form identified with ψ. This section of N vanishes where ψ, θ = 0 for θ `m. Conversely, those which vanish somewhere in `m are h i ∈ ∗ the set of decomposable elements (V+ V−)m, i.e. elements of rank equal to ⊗ 1 or less than 1 in Hom(V+,V−)m. If α (Λc )m annihilates a spinor ψ, then 0 = α α ψ = (α, α)ψ and α has length∈ 0, and conversely, is α has length 0 then· its· rank is− less than or equal to 1.
7. We have a quaternionic structure on V−, i.e. a function j : V− V− such 2 → that j = Id. Therefore, the restriction of j in P (V−), τ = jP (V−) is an − 2 antiholomorphic involution, τ = Id. On each fibre CP1, i.e. each twistor line `m, τ acts as the antipodal map taking the fibre as S2. P (V−) has no fixed points but leaves the fibres invariant.
Let us finish this section with a theorem that gives the two main properties that define the twistor space of a self-dual manifold. The previous remarks prove the following:
Theorem 3.4. Let be the twistor space of a self-dual manifold M, i.e W− = 0. Then [4]: Z
1. The fibres of π : M are holomorphic curves in . Each is a rational Z → Z curve CP1 whose normal bundle in is isomorphic to (1) (1) Z O ⊕ O 2. possesses a free antiholomorphic involution τ : which transforms Zeach fibre to itself. Z → Z
19 Z
T (CP1)= (2) O
b
1 π− (m)= ℓm ∼= CP1
π
b m M
Figure 1: Diagram of twistor space
3.4 Properties of the twistor space of a self-dual manifold
Let us study now properties of the 3-complex manifold , the twistor space of Z a self-dual manifold M. For each point m, `m denotes the fibre at this point, the twistor line. A vector v TmM defines a vectorv ˜ of the normal bundle ∈ N = T ` /T `m whose value at a point z `m is given by the inverse of the Z| m ∈ mapping π∗z : Tz TmM evaluated at v. Furthermore, if a conformal structure is defined on M,Z a → horizontal distribution is defined and the sectionv ˜ can be g H g −1 interpreted as a sectionv ˜ T `m , wherev ˜ (z) = (π∗z Hz ) (v). Most of the statements of the next theorem∈ followsZ| from the remarks of| the previous section:
Theorem 3.5. If the conformal structure on M is self-dual, then [23]
0 1. The sections v˜ are in fact holomorphic, i.e v˜ H (`m,N). Furthermore, g ∈ 0 given a metric in the conformal class of M, v H (`m,T ` ). Thus, H` ∈ Z| m m is a holomorphic sub-bundle of T ` isomorphic to N. Z| m 2. The real structure i on induces a real structure (C-antilinear involution) i∗ 0 Z on H (`m,N), which leaves the sections v˜ invariant. Conversely, if a section 0 σ H (`m,N) is real invariant then σ =v ˜ for some v TmM. ∈ ∈ 3. The complexification of the linear correspondence v v˜ induces an iso- 0 → morphism (TcM)m = H (`m,N) that preserves the respective real structures. ∼ 0 Thus, TmM is a real form of H (`m,N).
0 4. Every non-zero section σ H (`m,N) vanishes in at most one point of `m. ∈ Sections coming from TmM are nowhere vanishing.
0 5. The metric g on TmM induces a quadratic form g˜ on H (`m,N) whose null 0 cone consist of the sections σ H (`m,N) that vanish at some point of `m. ∈
20 Taking N = (1) (1) as the normal bundle for each twistor line `m, we have 1 O ⊕ O that H (`m,N) = 0 (see [13, page 92]). The twistor lines are compact complex submanifolds. Due to the theorem proved by Kodaira [14], there exists a complex manifold M c that parametrizes the twistor lines and there exists a canonical iso- c 0 c morphism TpM ∼= H (`p,N), where p M and `p is the corresponding twistor line. Due to this isomorphism, M c has a∈ complex dimension 4. The antipodal map c gives, as previously mentioned, a real structure to P (V−) but also to M . The fixed point set of M c is M. We can also derive from the remarks of the previous section that the cone of c 1 null vectors in M is in correspondence with sections of H (lp,N) that vanishes at some point. Taking that N = (1) (1) = (aµ + b, cµ + d), a section vanishes at some point if and only if ad Ocb =⊕ 0. O Therefore, we can say that a vector on the tangent space of M c is null if− the corresponding section obeys the previous quadric form. This null cone is real and defines a conformal structure on M c and also on M using the real structure. A metric within this conformal structure can be fixed afterwards by a volume form. Let us focus now on the case where M is also Ricci flat, thus being an Einstein manifold. The connection on V− induced by the Levi-Civita connection is also flat. If M is simply connected we can use parallel transport to identify all the fibres of M in a way that does not depend on the path. Moreover, if V− is flat, the identification under a closed path is the identity. We have then a fibration π : CP1. If Z → one considers the tangent space at a point z , Tz = Tπ(z)CP1 Nπ(z) = ∈ Z Z ∼ ⊕ ∼ (2)π(z) (1)π(z) (1)π(z), using that T CP1 = (2). Hence, the dual space O ⊕ O ⊕ O ∼ O is: T ∗ ( 2) ( 1) ( 1) . Recalling that K = V3 T ∗ is the z ∼= π(z) π(z) π(z) canonicalZ bundleO − of ⊕, then O −19; ⊕ O − Z Z 3 ^ ∗ Kz = ( ( 2)π(z) ( 1)π(z) ( 1)π(z)) = ( 4)π(z) K = π ( 4) ∼ O − ⊕ O − ⊕ O − ∼ O − ⇒ ∼ O − The bundle Hom(π∗ ( 4),K) is isomorphic to H0(π∗ (4) K) and a non-vanishing section of this vectorO bundle− is identified with an isomorphismO ⊗ in Hom(π∗ ( 4),K). Hence, the twistor space of a Ricci flat manifold M must have a non-vanishingO − sec- tion of K π∗ (4). This condition is also sufficient20, and will be used in the next section. ⊗ O
19This is also proved in [13, page 92]. 20This is proved in [4, page 390].
21 4 Construction of Gravitational Instantons
The aim of this section is to construct gravitational instantons using the tools of the previous section. For this purpose we will use the Reverse Penrose construction. First of all, a brief introduction to gravitational instantons is presented. Afterwards, the ideas introduced by Hitchin at [12] are explained.
4.1 Gravitational Instantons
Classical vacuum Einstein equations derive from the Euclidean gravitational action: Z Z 1 1 4 1 1 3 S[g] = (R 2Λ) g 2 d x [K]h 2 d Σ (4.1) −16πG M − − 8πG ∂M where Λ is the cosmological constant, R is the Ricci scalar and the second integral is the boundary term that will be described latter. Gravitational instantons could be defined to be complete non-singular positive definite solutions of the classical vacuum Einstein equations, either with a Λ term or without it. The property of self-duality is sometimes included by some authors into the previous definition (because of the notion of instanton). Some different topolog- ical properties can be found among these solutions which let classify them. There are solutions which can be considered the closest analogues to Yang-Mills instan- tons. These metrics are non-compact solutions with Λ = 0 and are asymptotically flat in all four dimensions. The Euclidean gravitational action (4.1) is not invariant under conformal trans- formations. As a result, since any asymptotically Euclidean metric must be an ex- tremum, it will have zero action. Due to the Positive Action Theorem [20] (which states that any asymptotically Euclidean self-dual metric has positive action except for the flat space which has zero action), there won’t be any self-dual asymptotically Euclidean metrics. The latter theorem does not exclude working with Asymptotically Locally Eu- clidean (ALE) spaces. These solutions tend to a flat metric modulo identification under a discrete subgroup of SO(4). This solutions are Ricci flat. The objective of this section is obtain these last solutions using twistor theory. We will work for the case when the topology of the manifold at infinity is S3/Γ where Γ is a cyclic subgroup of SU(2) acting on the unit sphere S3 in C2.
4.2 Hitchin’s Construction
The aforementioned cyclic subgroup acting on C2 is represented by the set of ma- trices: 2πin ! e k 0 − 2πin , 0 n k 0 e k ≤ ≤
22 We can see that the origin of the quotient space C2/Γ contains a singularity. As it is commonly done in algebraic geometry, affine algebraic varieties can be represented 2 in terms of the ring of functions from the variety itself to a field. If (z1, z2) C , k k ∈ the monomials z1 , z2 , z1z2 are invariant under left multiplication by elements of Γ. k k k If we take x = z1 , y = z2 , z = z1z2, they satisfy xy = z . This gives an isomorphism 2 3 k between C /Γ and (x, y, z) C xy = z , i.e. the set of solutions of the complex ∈ | surface xy zk = 0 in C3. − We can use now deformation theory to solve this singularity. One way to make a deformation is by adding lower terms in z and obtaining a family of deformations that also have the topology of S3/Γ at infinity:
k k−1 xy = z + a1z + ak ··· If the discriminant of the right hand side is non-zero for a particular selection of the lower terms , the resulting manifold has no singularity. This occurs because of the regular value21 theorem22. Therefore, we would take M to be the underlying 4-dimensional differentiable manifold of such a surface. In the previous section the relation between self-dual manifolds and complex geometry has been explained. This reformulation of the self-dual Einstein equations means that in order to find these gravitationals instantons it is necessary to look for the following:
(1) A complex 3-manifold , the total space of a holomorphic fibration π : Z Z → CP1 (2) A 4-parameter family of sections, each with normal bundle N = (1) (1) O ⊕ O (3) A non-vanishing holomorphic section θ of K π∗ (4) ⊗ O (4) A real structure on such that π and s are real, and is fibred by the real Z Z sections of the family. Here CP1 is given the real structure of the antipodal map, i.e u u¯−1 → − In order to construct we need a family of complex structures on M which are Z parametrized by u CP1. The fibre of each u CP1 defines a curve in (where each point of the curve∈ is parallel transported from∈ u) which is also a sectionZ from M to . This section defines a complex structure on M as we have seen in the previousZ section. We can say then that we can construct a family of complex structure parametrized by u. We call ¯ the first approximation of and is defined by: Z Z k k−1 xy = z + a1(u)z + ak(u) (4.2) ··· where ai(u) is locally a holomorphic function of u. We emphasize the term locally because CP1 is a compact complex manifold and therefore global functions are 21see Definition A.6.6. 22If the discriminant is not zero, all the roots are different and then the differential of xy zk k 1 k k 1 − − a z − a is not 0 at the roots. Therefore, the value 0 of the function xy z a z − a 1 −· · ·− k − − 1 −· · ·− k is a regular value and the pre-image is a smooth submanifold of C3.
23 constant functions. Vector bundles of dimension 1, i.e line bundles, on CP1 are isomorphic to (k) for k Z. Therefore, we take ai(u) as a line bundle, i.e. ai(u) (n i) and itO implies x ∈ (l), y (m), z (n) where l + m = kn in order for∈ theO equation· to make sense.∈ O Hence,∈ O¯ is a 3-dimensional∈ O hyper-surface solution of the aforementioned equation in the complexZ 4-manifold (l) (m) (n). Each O ⊕ O ⊕ O u CP1 will then define the 2-dimensional complex manifold. ∈ ¯ 2 We need a real structure on . The map on C defined by (z1, z2) ( z¯2, z¯1) Z → − induces a real structure on CP1, which is the aforementioned antipodal map σ. Since Γ acts by left multiplication, it conmutes with the real structure and it induces also a real structure on C2/Γ given by: x ( 1)ky,¯ y x,¯ z z¯. We can use this to define a real structure on (l) →(m)− (n): → → − O ⊕ O ⊕ O τ(x, y, z, u) = ( 1)kσ(y), σ(x), σ(z), σ(u) − y¯ x¯ z¯− = ( 1)k , , , u¯−1 − u¯k u¯k −u¯2 − where σ : (m) (m) is the antipodal map defined before. This real structure O → O ¯ requires that l = m. If the sections are also real, i.e. σ(ai(u)) = ai(σ(u)), then ¯ Z has a real structure and the fibration is also real π : CP1. Z → At this point, we have obtained a real structure for ¯ and a real projection Z to CP1. However, as we have explained before, at the points uk CP1 where the discriminant is zero the resultant is not a manifold. To obtain a∈ non-singular ¯ manifold from we can resolve these singularities by taking a finite covering of CP1 Z branched over those points uk. That would imply that the base Riemann manifold where the new non-singular manifold is fibred is of higher genus23. However, taking the assumption that Z
k k X k−i Y n ai(u)z = (z pj(u)) pj Γ(H ) i=0 j=1 − ∈ we obtain a complex 3-manifold with a real structure and a compatible fibration Z π : CP1. So, is not fibred over another Riemannian surface but over CP1. TheZ reason → for thisZ is found in [5] . In order to determine the values of l, m, n, we use the previous condition l = m, l + m = kn because of the compatibility with (4.2) and finally we use now the aforementioned condition (2). The normal bundle of CP1 can be identified as the tangent space of the fibres, i.e. with the tangent space of each π−1(u). If we take a section which misses the singularity, the tangent space of the fibres is the kernel of this vector bundle homomorphism
(fx, fy, fz): (l) (m) (n) (kn) O ⊕ O ⊕ O → O This is also due to the regular value theorem: all the elements contained in the submanifold have as image through f the regular value mentioned before, hence
23The genus g of a closed Riemann surface can be defined in terms of the Euler characteristic, χ = 2 2g. − 24 the tangent vectors on this manifold are in the kernel of df. Using the first Chern class, we know that c1(N) = c1( (1) (1)) = 2 and in our case, c1(N) = l + m + n kn = n. Therefore, n =O 2 and⊕ Ol = m = k. −
Z
b b
TCP1 b NCP1
CP1 b b
b b M
Figure 2: Projection to CP1 We also need a non-vanishing section of the vector bundle K π∗ (4) on because of condition (3) . Let us make the assumption that a constant⊗ O multipleZ of the standard form is non-vanishing on : Z dx dy dz dy dz du dx dz du dx dy du ∧ ∧ = ∧ ∧ = ∧ ∧ = ∧ ∧ fu fx fy fz Once we will have the metric this fact will be noticeable from its existence.
Finally, let us see how the sections of this vector bundle on CP1 are parametrized by a 4-dimensional complex manifold as Kodaira’s theorem implies. Through this identification we define then sections of (k) (k) (2). A section of (m) can be defined by a polynomial of complexO coefficients⊕ O ⊕ of O degree m in u (seeO [13, page 91]) . Hence, we take polynomials x(u), y(u), z(u) such that:
k Y x(u)y(u) = (z(u) pj(u)) j=1 −
2 2 where z(u) = au + 2bu + c and pi = aiu + 2biu + ci. Taking αi and βi the roots of the polynomial z(u) pi(u) and doing a simple factorization we obtain: − k k Y Y x = A (u αi) , y = B (u βi) i=1 − i=1 −
25 Q where AB = (a ai). At this point, the curves are parametrized locally by (a, b, c, A) where the− four of them are complex numbers. Let us see how the real lines are parametrized. To obtain real lines we must impose that they are compatible with the real structure. Therefore, as it is seen before
x(σ(u)) = ( 1)kσ(y(u)), y(σ(u)) = σ(x(u)), z(σu) = σ(z(u)) − − and the third condition implies that c = a,¯ b = ¯b: − z(σu) = au¯−2 2bu¯−1 + c − a¯u¯2 + 2bu¯ + c σ(z(u)) = = cu¯−2 2bu¯−1 a¯ − − u¯2 − − − Therefore, the roots are:
p 2 p 2 2 2(b bi) 4(b bi) 4(a ai)(c ci) (b bi) (b bi) + a ai − − ± − − − − = − − ± − | − | 2(a ai) a ai − − 2 Calling ∆i the real positive square root of (b bi) (a ai)(c ci), then − − − − k k Y (b bi) ∆i Y (b bi) + ∆i x = A u + − − , y = B u + − (4.3) a ai a ai i=1 − i=1 −
Finally, imposing again the reality condition on x(u) and y(u) and that AB = Q (a ai): − ¯ Y AA = ((b bi) + ∆i) (4.4) − Therefore, A 2 = AA¯ can be described in terms of a, b and we only need its angle to define A|completely.| Therefore, with the real variables Re(a), Im(a), b and the angle arg(A) we can parametrize the real lines. We still have the indeterminacy of the polynomials pi, i 1 i k, therefore each metric is determined by fixing the ∈ ≤ ≤ coefficients, which is the same as taking k points in R3. Taking a general point (a, b, c, A) of the manifold which parametrizes the sec- tions in ¯, i.e. the manifold M c, we obtain a section x(u), y(u), z(u) by Kodaira’s theorem.Z This theorem also associated each tangent vector of the point (a, b, c, A) to a section of the normal bundle of the associated section x(u), y(u), z(u). We will call (a0, b0, c0,A0) the tangent vector and x0(u), y0(u), z0(u) the section of the normal bundle. We can now define the null cone on M c as the sections of the normal bundle which vanish at some point. Therefore, we need a common zero for x0(u), y0(u), z0(u): z0(u) = a0u2 + 2b0u + c0 = 0 (4.5) x0(u) A0 X α0 = i = 0 (4.6) x(u) A − u αi − x0(u) 0 Q 0 where the second equation follow from: x(u) = ln(x(u)) = ln (A (u αi)) = 0 0 − P 0 A P αi (ln A + ln(u αi)) = . − A − u−αi
26 c−ci On the other hand, using αiβi = and that z pi = (a ai)(u αi)(u βi) we a−ai obtain: − − − − −1 z pi = (u αi) (a ai)u (c ci)α − − − − − and differentiating it gives:
0 0 −1 0 0 −1 −2 0 0 = z (u) = α (a ai)u (c ci)α ) + (u αi)(a u c α + (c ci)α α ) = − i − − − i − − i − i i 0 −1 −2 0 0 −1 = α (a ai)u + (c ci)α + (u αi)(c ci)α + (u αi)(a u c α ) i − − − i − − i − − i ⇒ 0 0 0 0 0 α a uαi c αia u c i = − = − ⇒ u αi (a ai)u(αi βi) 2u∆i − − − 2 1/2 where 2∆i = (a ai)(αi βi) = 2(b ac) . Once we obtain− this result,− we substitute− it in (4.6):
0 0 0 0 0 A X αia u c 2A X αia X c = − u 0 + = 0 A 2u∆i ⇒ A − ∆i ∆i
Changing the variables γ = P 1 , δ = P αi : ∆i ∆i 2A γc0 u δa0 + γc0 = 0 u = − A0 − ⇒ 2A δa0 A0 − and substituting in (4.5):
γ2c02 γc0 2A 2A 2 a0 +2b0 − +c0 = 0 γ2a0c0 2γb0 δa0 + δa0 = 0 2A 02 2A 0 0 0 δa 0 δa ⇒ − A − A − A0 − A − 0 2A 02 where we have divided the entire equation by c and multiplied by A0 δa . Expanding now the equation: − 4A0 4A02 4A0 γ2a0c0 γb0 + 2δa0γb0 + + δ2a02 δa0 = 0 − A A2 − A Now, adding and subtracting γ2b02, we obtain:
2A0 2 γ2(a0c0 b02) + δa0 γb0 = 0 − A − − which defines the null cone. At this point , we have considered a general point curve x(u), y(u), z(u). The computations for real lines made before gives:
0 0 0 2A X (b bi) + ∆ Re = − i = γb0 + Re(δa0) A (b bi) + ∆i − Let us see this equality. From (4.4):