DEFORMATION THEORY OF NEARLY KÄHLER MANIFOLDS LORENZO FOSCOLO Abstract. NearlyKählermanifoldsaretheRiemannian6–manifoldsadmittingrealKillingspinors. 6 Equivalently, the Riemannian cone over a nearly Kähler manifold has holonomy contained in G2. In this paper we study the deformation theory of nearly Kähler manifolds, showing that it is ob- 1 0 structedingeneral.Moreprecisely,weshowthattheinfinitesimaldeformationsofthehomogeneous 2 nearly Kähler structureon theflag manifold are all obstructed to second order. n a 1. Introduction J A Killing spinor on a Riemannian spin manifold (Mn,g) is a spinor ψ such that 8 1 ∇ ψ = αX ·ψ, X ] for some α ∈ C and all vector field X. Here · denotes Clifford multiplication. Killing spinors G appeared in work of Friedrich [8] on the first eigenvalue of the Dirac operator. It was shown in [8] D that every manifold with a Killing spinor is Einstein: Ric(g) = 4(n−1)α2g. In particular, one of . h three cases must hold: (i) α 6= 0 is purely imaginary and M is non-compact; (ii) if α = 0 then ψ t is a parallel spinor and therefore (M,g) has holonomy contained in SU(n), Sp(n), G or Spin(7); a 2 4 2 m (iii) α 6= 0 is real: ψ is called a real Killing spinor and M (if complete) is compact with finite [ fundamental group. In the real case one can always assume that α= ±1 by scaling the metric. 2 There is in fact a relation between parallel spinors and real Killing spinors: by work of Bär [2], 1 v the cone over a manifold with a real Killing spinor has a parallel spinor and conversely simply 0 connected manifolds with a real Killing spinor are the cross-sections of Riemannian cones with 0 holonomy contained in SU(n), Sp(n), G or Spin(7), depending on the dimension n and the 4 2 4 2 number of linearly independent real Killing spinors. 4 0 Nearly Kähler manifolds are the 6–dimensional Riemannian manifolds admitting real Killing . spinors. The cone over a nearly Kähler manifold has holonomy contained in G . 1 2 0 Remark 1.1. ThenamenearlyKählerwasintroducedbyGray[10]todenoteaspecialclassofalmost 6 1 Hermitian manifolds in every even dimension. What we call here nearly Kähler manifolds are often : referred to as strict nearly Kähler manifolds of dimension 6. The terminology Gray manifolds has v i also been used, cf. [15, Definition 4.1]. X r Despite thespinorial pointof view will play arole in this paper,we prefer to relate theholonomy a reduction of the cone C(M) over a nearly Kähler manifold to the existence of a closed and co- closed stable 3–form rather than to the existence of a parallel spinor. From this point of view a nearly Kähler structure is an SU(3) structure with special torsion: a pair of differential forms (ω,Ω), where ω is a non-degenerate 2–form and Ω is a complex volume form satisfying appropriate algebraic compatibility conditions and the first order PDE system dω = 3ReΩ, dImΩ= −2ω2. There are currently only six known examples of simply connected nearly Kähler manifolds. Four of these are homogeneous and were known since 1968 [22]: the round 6–sphere endowed with the non-integrable almost complex structure induced by octonionic multiplication on R7 ≃ ImO and the 3–symmetric spaces CP3 = Sp(2)/U(1)×Sp(1), S3×S3 = SU(2)3/SU(2) and F = SU(3)/T2. 3 Recently two inhomogeneous nearly Kähler structures on S6 and S3×S3 were found in [7]. Finite quotients of the homogeneous nearly Kähler manifolds have also been studied [6]. This scarcity of 1 2 L.FOSCOLO examples should be contrasted with the infinitely many known examples of manifolds with real Killing spinors in other dimensions: Sasaki–Einstein, 3–Sasaki and nearly parallel G manifolds 2 (the cross-sections of Calabi–Yau, hyperkähler and Spin(7) cones, respectively). In this paper we study the deformation theory of nearly Kähler manifolds. In [15] Moroianu– Nagy–SemmelmannstudiedinfinitesimaldeformationsofnearlyKählerstructures,identifyingthem with an eigenspace of the Laplacian acting on coclosed primitive (1,1)–forms. The question of whethernearlyKähler6–manifoldshavesmooth,unobstructeddeformationswashowever leftopen. Because of thescarcity of examples itwouldbevery interesting tounderstandwhetheritispossible to obtain new nearly Kähler manifolds by deforming known ones. Understanding whether nearly Kähler deformationsareingeneral obstructedis alsoimportantforapplications tothetheoryof G 2 conifolds (asymptotically conical and conically singular G manifolds) developed by Karigiannis– 2 Lotay [12]. Finally, possible constructions of new examples of nearly Kähler manifolds based on singular perturbation methods require as a preliminary step the study of the deformation theory of nearly Kähler manifolds (and its extension to certain singular nearly Kähler spaces). ItisinstructivetorecallknownresultsaboutthedeformationtheoryofmanifoldswithrealKilling spinors in other dimensions.Continuous families of Sasaki–Einstein structures are certainly known, e.g. the regular Sasaki–Einstein 5–manifolds obtained from del Pezzo surfaces of degree d ≤ 4 via the Calabi ansatz have non-trivial moduli. However in general Sasaki–Einstein manifolds have obstructed deformations (cf. [20] forthe relation between integrability of infinitesimal deformations and K–stability in the more general context of constant scalar curvature Sasaki metrics). By a result of Pedersen–Poon [19] 3–Sasaki manifolds are rigid. In [1] Alexandrov–Semmelmann study infinitesimal deformations of nearly parallel G structures. As in the nearly Kähler case these are 2 identified with a certain subspace of an eigenspace of the Laplacian acting on 3–forms. It is unclear whether infinitesimal deformations of nearly parallel G manifolds are unobstructed in general. 2 Given what is known about deformations of Sasaki–Einstein manifolds, it is not surprising that nearly Kähler 6–manifolds have obstructed deformations in general, as we show in this paper. In [13, Theorem 6.12] Koiso showed that infinitesimal deformations of Einstein metrics are in general obstructed. He exhibited Einstein symmetric spaces with non-trivial infinitesimal Einstein deformations which cannot be integrated to second order. We will follow a similar strategy. In [16] Moroianu–Semmelmann calculated the space of infinitesimal deformations of the homo- geneous nearlyKählermanifolds.TheyfoundthatCP3 andS3×S3 arerigidwhiletheflagmanifold F has an 8–dimensional space of infinitesimal deformations [16, Corollary 6.1]. (The case of the 3 round 6–sphere is special, since there are more than two Killing spinors in this case. The space of nearly Kähler structures compatible with the round metric is an RP7–bundle over S6. Since the round metric does not admit any Einstein deformation, there are no infinitesimal nearly Kähler deformations other than the ones coming from this family.) In this paper we address the question of deciding whether the homogeneous nearly Kähler struc- ture on the flag manifold admits genuine nearly Kähler deformations. Theorem. The infinitesimal nearly Kähler deformations of the flag manifold F are all obstructed. 3 TheproofoftheTheorem(andthepaperitself)isdividedintotwodistinctparts.Inafirststepwe will obtain a deeper understandingof the deformation theory of nearly Kähler structures in general beyond the infinitesimal level considered in [15]. The main tool is a certain Dirac-type operator on nearly Kähler manifolds, which appears as a certain combination of differential, codifferential and type decomposition acting on differential forms. The use of Dirac operators as tools in the deformation theory of manifolds with special geometric structures is not new. Nördstrom [17, Chapter 3] used Dirac operators to streamline certain steps in the deformation theory of manifolds with special holonomy. More precisely, mapping properties of Dirac operators are used to establish slice theorems for the action of the diffeomorphism group. This approach has turned out to be particularly useful in the deformation theory of non-compact manifolds with special holonomy, in particular asymptotically cylindrical manifolds [18] and conifolds [12]. DEFORMATION THEORY OF NEARLY KÄHLER MANIFOLDS 3 Besides the application to the deformation theory of nearly Kähler structures, we will show that Dirac-type operators can also be used to obtain interesting results about Hodge theory on nearly Kähler manifolds.In particular, we willgive an elementary proof of a resultof Verbitsky [21, Theorem 6.2 and Remark 6.4] on the type decomposition of harmonic forms on nearly Kähler manifolds. A second important ingredient in our treatment of the deformation theory of nearly Kähler manifolds is Hitchin’s interpretation of nearly Kähler structures as (constrained) critical points of a Hamiltonian function on the infinite dimensional symplectic vector space Ω3 ×Ω4 . This exact exact description will allow us to interpret the nearly Kähler equations as the vanishing of a smooth map Φ: Ω3 ×Ω4 ×Ω1 → Ω3 ×Ω4 , rather than as equations on the space of SU(3) exact exact exact exact structures. The main advantage of this approach is to introduce additional free parameters that can be used to reduce the cokernel of the linearisation DΦ and pin down exactly where possible obstructions to integrate infinitesimal deformations could lie. The general deformation theory of nearly Kähler structuresis then applied to the specific case of the flagmanifold F .Theframework introduced in thefirstpartof the paperwill make possiblethe 3 explicit calculation of the non-vanishing obstructions to integrate the infinitesimal deformations of the homogeneous nearly Kähler structure on F to second order. 3 As we have already mentioned, Alexandrov–Semmelmann [1] studied infinitesimal deformations of nearly parallel G manifolds. In particular, by [1, §8] the normal homogeneous nearly parallel 2 G manifolds are all rigid except for the Aloff–Wallach manifold SU(3)/U(1), which admits an 2 8–dimensional space of infinitesimal deformations isomorphic to su . It is likely that the methods 3 of this paper could also be used to analyse the integrability of these infinitesimal deformations. The paper is organised as follows. In Section 2 we collect various preliminary results about 6– manifolds endowed with an SU(3) structure. The notion of a stable form introduced by Hitchin in [11] will be central in our exposition. These results are known and we collect them here simply for the convenience of the reader. In Section 3 we study algebraic compatibilities on nearly Kähler manifolds between differential and codifferential and the decomposition of the space of differential forms into types corresponding to irreducible representations of SU(3). We will then introduce the Dirac-type operator mentioned above and study its mapping properties. As a first application, we will derive results about the Hodge theory of nearly Kähler manifolds. Section 4 discusses the deformation problem of nearly Kähler manifolds. The Dirac-type operator is used to define a slice fortheactionofthediffeomorphismgroupwhileHitchin’sinterpretationofnearlyKählerstructures as (constrained) critical points allows to re-write the nearly Kähler equations as the vanishing of a certain non-linear map Φ: Ω3 ×Ω4 ×Ω1 → Ω3 ×Ω4 . We study the linearisation DΦ at exact exact exact exact a nearly Kähler structure and identify its cokernel and therefore possible obstructions to integrate infinitesimaldeformationstogenuinenearlyKählerdeformations.Deformationsofthehomogeneous nearlyKählerstructureontheflagmanifoldarefinallystudiedinSection5.Weintroducethenotion of second order deformations, recall the result of Moroianu–Semmelmann [16, Corollary 6.1] on the existence of an 8–dimensional space of infinitesimal nearly Kähler deformations and, via explicit calculations and representation theoretic considerations, we show that these are all obstructed to second order. Acknowledgements. The author wishes to thank Johannes Nordström and Uwe Semmelmann forinterestingconversations relatedtothispaperandUweSemmelmannforcommentsonanearlier draftofthisnote.PartofthisworkwascarriedoutwhiletheauthorwasvisitingLeibnizUniversität Hannover in the Fall 2015; he wishes to thank the Riemann Center for Geometry and Physics for support and the Differential Geometry Group for hospitality. 2. SU(3) structures on 6–manifolds InthispreliminarysectionwecollectvariousknownfactsaboutSU(3)structureson6–manifolds. These results are well-known and we collect them here simply for the convenience of the reader. 4 L.FOSCOLO 2.1. Stable forms. Following Hitchin [11], the notion of a stable form will be central in our exposition. Definition 2.1. A differential form φ∈ Λp(Rn)∗ is stable if its GL(n,R)–orbit in Λp(Rn)∗ is open. In dimension 6, there are only three possibilities for stable forms [11, §2]: (i) astable2–formω(anon-degenerate2–form)withopenorbitisomorphictoGL(6,R)/Sp(6,R); (ii) a stable 4–form σ, with stabiliser Sp(6,R); (iii) a stable 3–form ρ, with stabiliser SL(3,C). Note that in all three cases the stabiliser is in fact contained in SL(6,R) and therefore to each stable form one can associate a volume form dv. For example one can define dv(ω) = 1ωn for a n! stable 2–form ω in dimension 2n. Using the homogeneous behaviour of the map dv, for every stable p–form φ Hitchin defines its dual φˆ, a (n−p)–form such that dv(φ) is proportional to φ∧φˆ. In dimension 6 we have: (i) for a stable 2–form ω, ωˆ = 1ω2; 2 (ii) for a stable 4–form σ, σˆ is the unique non-degenerate 2–form such that σ = 1σˆ2; 2 (iii) for a stable 3–form ρ, ρˆis the unique 3–form such that ρ+iρˆis a nowhere vanishing complex volume form. Inparticular,indimension6therealpartofacomplex volumeformΩuniquelydeterminesΩitself. Moreover, since its stabiliser is SL(3,C), a stable 3–form ReΩ defines an almost complex structure J: a 1–form α is of type (1,0) if and only if α∧Ω = 0. Definition 2.2. An SU(3)–structure on a 6–manifold M is a pair of smooth differential forms (ω,ReΩ) such that (i) ω is a stable 2–form with Hitchin’s dual 1ω2; 2 (ii) ReΩ is a stable 3–form with Hitchin’s dual ImΩ; (iii) the following algebraic constraints are satisfied: (2.3) ω∧ReΩ = 0, 1ω3 = 1ReΩ∧ImΩ. 6 4 The two constraints guarantee that the stabiliser of the pair (ω,ReΩ) is exactly SU(3) = Sp(6,R)∩SL(3,C). We could of course define an SU(3)–structure as the choice of a pair of stable differential forms σ ∈ Ω4(M) and ReΩ ∈Ω3(M) satisfying (2.3) with ω = σˆ. On R6 ≃ C3 with holomorphic coordinates (z ,z ,z ) we define the standard parallel SU(3)– 1 2 3 structure (ω ,ReΩ ) by 0 0 ω = i (dz ∧dz +dz ∧dz +dz ∧dz ), Ω = dz ∧dz ∧dz . 0 2 1 1 2 2 3 3 0 1 2 3 An SU(3) structure on a 6–manifold M in the sense of Definition 2.2 defines a reduction of the structure group of the tangent bundle of M to SU(3) by considering the sub-bundle of the frame- bundle of M defined by {u: R6 −∼→ T M|u∗(ω ,ReΩ ) = (ω ,ReΩ )}. x x x 0 0 Remark 2.4. Since SU(3) ⊂ SU(4) ≃ Spin(6) is precisely the stabiliser of a non-zero vector in C4, we could also define an SU(3) structure as the choice of a spin structure on M together with a non-vanishing spinor. Note that everySU(3)–structure induces aRiemannian metricg becauseSU(3) ⊂ SO(6). Hence from now on we will identify without further notice TM and T∗M using the metric g (and R6 with (R6)∗ using the flat metric g ). 0 Remark 2.5. In particular, when we write JX and think of it as a 1–form we really mean (JX)♭. The fact that (JX)♭ = −JX♭ might cause some confusion at times. We will instead distinguish between differential df and gradient ∇f of a function f. DEFORMATION THEORY OF NEARLY KÄHLER MANIFOLDS 5 2.2. Decomposition of the space of differential forms. The decomposition into irreducible representations of the SU(3)–representation Λ∗R6 is well-known. This is usually stated after com- plexification in terms of the (p,q)–type decomposition induced by the standard complex structure J andin terms of primitiveforms.We willstick with realrepresentations andfindmoreconvenient 0 to use the uniform notation Λk for an irreducible component of ΛkR6 of dimension ℓ. ℓ Lemma 2.6. We have the following orthogonal decompositions into irreducible SU(3) representa- tions: Λ2R6 = Λ2⊕Λ2⊕Λ2, 1 6 8 where Λ2 = Rω, Λ2 = {XyReΩ|X ∈R6} and Λ2 is the space of primitive forms of type (1,1). 1 6 8 Λ3R6 = Λ3⊕Λ3 ⊕Λ3 , 6 1⊕1 12 where Λ3 = {X ∧ω|X ∈ R6}, Λ3 = RReΩ⊕RImΩ and Λ3 is the space of primitive forms of 6 1⊕1 12 type (1,2)+(2,1), i.e. Λ312 = {S∗ReΩ|S ∈ Sym2(R6),SJ +JS = 0}. Remark 2.7. HereanendomorphismS ∈End(R6)actsonadifferentialp–formφbyS∗φ(X1,...,Xp) = p − φ(X ,...,SX ,...,X ). j=1 1 j p PFor a 6–manifold M endowed with an SU(3)–structure (ω,Ω) we denote by Ωk(M) the space of ℓ smooth sections of the bundle over M with fibre Λk. ℓ By definition Λ2 is isomorphic to Λ1 via the metric and the contraction with ReΩ. The adjoint 6 of this map with respect to the flat metric g will be denoted by α: Λ2 → Λ1. Note that for a form 0 6 η ∈ Λ2 we have 6 (2.8) η = 1α(η)yReΩ. 2 The following identities follow from [15, Equations (12), (17), (18) and (19)]. Lemma 2.9. In the decomposition of Lemma 2.6 the Hodge–∗ operator is given by: (i) ∗ω = 1ω2; 2 (ii) ∗(XyReΩ)= −JX ∧ReΩ = X ∧ImΩ; (iii) ∗(η ∧ω)= −η for all η ∈ Ω2; 0 0 0 8 (iv) ∗(X ∧ω)= 1Xyω2 = JX ∧ω; 2 (v) ∗ReΩ = ImΩ and ∗ImΩ = −ReΩ; (vi) ∗(S∗ReΩ)= −S∗ImΩ =(JS)∗ReΩ; We use Lemma 2.9 to deduce useful identities and characterisations of the different types of forms. Lemma 2.10. If η = η +λω+XyReΩ ∈ Ω2 with η ∈ Ω2, then the following holds: 0 0 8 (i) ∗(η∧ω)= −η +2λω+XyReΩ; 0 (ii) ∗(η∧η∧ω)= −|η |2+6λ2+2|X|2; 0 (iii) ∗(η∧ReΩ)= 2JX and ∗(η∧ImΩ)= −2X; (iv) ∗(η∧ω2)= 6λ. In particular, η ∈ Ω2 iff η∧ω2 = 0 = η∧ReΩ iff η∧ω = −∗η. 8 Proof. (i) follows immediately from Lemma 2.9 (i)–(iii). In turn, (i) implies (ii) by the identity η∧η∧ω = η∧∗2(η∧ω), the fact that the decomposition of Lemma 2.6 is orthogonal and |ω|2 = 3, hXyReΩ,XyReΩi = 2|X|2. The identities (iii) follows immediately from [15, Equations (3) and (4)] and (iv) from 2∗ω = ω2 and |ω|2 = 3. (cid:3) We have similar identities on 3–forms, with analogous proof. Lemma 2.11. If σ = X ∧ω+λReΩ+µImΩ+S∗ReΩ ∈ Ω3, then the following holds: 6 L.FOSCOLO (i) ∗(σ ∧ω) = 2JX; (ii) ∗(σ ∧ReΩ)= −4µ; (iii) ∗(σ ∧ImΩ)= 4λ. In particular, ρ∈ Ω3 iff ρ∧ω = 0 = ρ∧Ω. 12 Finally, it will be useful to have an explicit formula for the linearisation of Hitchin’s duality map for stable forms in terms of the Hodge–∗ and the decomposition of forms into types. Proposition 2.12. Given an SU(3) structure (ω,ReΩ) on M6 let σ ∈ Ω4(M) and ρ∈ Ω3(M) be forms with small enough C0–norm so that 1ω2+σ and ReΩ+ρ are still stable forms. Decomposing 2 into types we write σ = σ1+σ6+σ8 and ρ= ρ6+ρ1⊕1+ρ12. (i) The image σˆ of σ under the linearisation of Hitchin’s duality map at 1ω2 is 2 σˆ = 1 ∗σ +∗σ −∗σ . 2 1 6 8 (ii) The image ρˆ of ρ under the linearisation of Hitchin’s duality map at ReΩ is ρˆ= ∗(ρ6 +ρ1⊕1)−∗ρ12. Proof. Inordertoprovethefirststatement,observethatσˆ istheunique2–formsuchthatσ = ω∧σˆ. Apply ∗ to this identity and use Lemma 2.10.(i). The formula for ρˆ follows from [15, Lemma 3.3] and the last three identities in Lemma 2.9. (cid:3) 2.3. Nearly Kähler manifolds. Given a subgroup G of SO(n), we define a G–structure on a Riemannian manifold (Mn,g) as a sub-bundle P of the orthogonal frame bundle of M with structure group G. Theintrinsic torsion of P is a measureof how much P is far from beingparallel with respect to the Levi–Civita connection ∇ of (M,g). More precisely, restricting ∇ to P yields a so(n)–valued 1–form θ on P. Choose a complement m of the Lie algebra of G in so(n). Projection of θ onto myields a 1–form T on M withvalues inthebundleP× m.Thisis the(intrinsic) torsion G of the G–structure P. For an SU(3)–structure on a 6–manifold M one can check that ∇(ω,Ω) = T∗(ω,Ω) where T acts on differential forms via the representation of m ⊂ so(6) on Λ∗(R6). It turns out that T itself is uniquely recovered by knowledge of the anti-symmetric part of T∗(ω,Ω), i.e. the knowledge of dω and dΩ. Proposition 2.13 ( [4, Theorem 2.9]). Let (ω,Ω) be an SU(3) structure. Then there exists func- tions w ,wˆ ∈ Ω0, w ,wˆ ∈ Ω2, w ∈Ω3 and vector fields w ,w on M such that 1 1 2 2 8 3 12 4 5 dω =3w ReΩ+3wˆ ImΩ+w +w ∧ω, 1 1 3 4 dReΩ = 2wˆ ω2+w ∧ReΩ+w ∧ω, 1 5 2 dImΩ = −2w ω2−Jw ∧ReΩ+wˆ ∧ω. 1 5 2 Note that the different sign in front of Jw ∧ReΩ in the formula for dImΩ with respect to the 5 formula in [4, Theorem 2.9] is consistent with Remark 2.5. Definition 2.14. An SU(3) structure on a 6–manifold M is called a nearly Kähler structure if wˆ ,w ,wˆ ,w ,w ,w all vanish and w = 1. In other words a nearly Kähler structure satisfies 1 2 2 3 4 5 1 (2.15) dω = 3ReΩ, dImΩ= −2ω2. As remarked in the Introduction, (2.15) are equivalent to the requirement that ϕ = r2dr∧ω+r3ReΩ is a closed and coclosed “conical” G structure on the cone C(M)= R+×M. Thus the cone C(M) 2 has holonomy contained in G and in particular is Ricci-flat. As a consequence, nearly Kähler 2 manifolds are Einstein with positive Einstein constant normalised so that Scal = 30. In particular every complete nearly Kähler manifold is compact with finite fundamental group. DEFORMATION THEORY OF NEARLY KÄHLER MANIFOLDS 7 In Remark 2.4 we observed that every 6–manifold with an SU(3) structure is spin and endowed with a unit spinor ψ. The nearly Kähler equations (2.15) have an equivalent interpretation as a first order differential equation on ψ [2, Theorem 2]. Indeed, every G manifold admits a parallel 2 spinor. Restricting the parallel spinor on the cone C(M) to the cross-section M induces a real Killing spinor ψ on every nearly Kähler manifold, i.e. (possibly after changing orientation) a unit spinor ψ such that (2.16) ∇ ψ = 1X ·ψ X 2 for every vector field X. Clifford multiplication by the volume form Vol (i.e. the complex structure on the spinor bundle) yields a second Killing spinor Vol·ψ satisfying (recall that X · Vol·ψ = −Vol·X ·ψ) ∇ (Vol·ψ) = −1X ·(Vol·ψ). X 2 In the rest of the paper (M,ω,Ω) will denote a complete (hence compact) nearly Kähler 6– manifold and ψ will denote the real Killing spinor on M satisfying (2.16). 3. Hodge theory on nearly Kähler manifolds ThemaingoalofthissectionistointroduceaDirac-typeoperatorD onanearlyKählermanifold M and study its mapping properties. D differs from the standard Dirac operator by a lower order term. This operator arises as a certain composition of differential, codifferential and type decom- position on differential forms and will play a central role in the study of the deformation theory of nearly Kähler manifolds. Furthermore, we find that the mapping properties of the operator D have interesting consequences on the Hodge theory of nearly Kähler manifolds. In particular, we will give an elementary proof of results of Verbitsky [21] on the type decomposition of harmonic forms on nearly Kähler manifolds. 3.1. Differential and co-differential on nearly Kähler manifolds. Before giving the defini- tion of the Dirac operator D we need to study how the exterior differential d: Ωk(M) → Ωk+1(M) and its adjoint behave with respect to the decomposition of Ω∗(M) introduced in Lemma 2.6. We need an additional piece of notation and few simple observations. LetΛ: Ωk(M) → Ωk−2(M)bethepoint-wiseadjointofwedgingwithωandnotethatΛ(X∧ω)= 2X for every vector field X. Recall also that α: Ω3(M) → Ω1(M) is the point-wise adjoint of the map X 7→ XyReΩ. Furthermore we have the following identities: (3.1) ∗(X ∧σ) = (−1)pXy∗σ for every p–form σ and X ∈ X(M); (3.2) (XyReΩ)∧ω = −JX ∧ReΩ, which follows immediately by contracting ReΩ∧ω = 0 by X. (3.3) (XyReΩ)∧ReΩ = X ∧ω2, (XyReΩ)∧ImΩ = JX ∧ω2 which are [15, Equations (3) and (4)]. (3.4) ∗Y = 1JY ∧ω2 2 which is a consequence of (3.1). As a consequence, (3.5) d∗Y = −∗(dJY ∧ 1ω2). 2 Proposition 3.6. For every f ∈ C∞(M), η ∈ Ω2(M) and X ∈ X(M) we have 8 (i) d(fω) = df ∧ω+3f ReΩ; (ii) d∗(fω)= Jdf; (iii) dη = 1Jd∗η∧ω+ρ for some ρ∈ Ω3 ; 2 12 (iv) d(XyReΩ)= − 1α(dJX)+3X ∧ω−1(d∗X)ReΩ−1d∗(JX)ImΩ+ρ′ for some ρ′ ∈ Ω3 ; 2 2 2 12 (cid:16) (cid:17) 8 L.FOSCOLO (v) d∗(XyReΩ)= Jα(dJX). Proof. (i) follows immediately from (2.15). Since ω2 is closed, d∗(fω)= −∗d(1fω2)= −∗(df ∧ 1ω2) and (3.4) completes the proof of (ii). 2 2 In order to prove (iii), differentiate η ∧Ω = 0 and use the fact that η ∧dΩ = −2iη ∧ω2 = 0. Lemma 2.11 then implies that dη has zero component in the complex line spanned by Ω. Similarly, differentiating the equality η∧ω+∗η = 0 and using Lemma 2.11.(i) yields Λdη = Jd∗η. The identity (iv) is proved in a similar way. Consider d(XyReΩ)∧ω: by (3.2) and (3.3) d(XyReΩ)∧ω = −3X ∧ω2−dJX ∧ReΩ. Lemma 2.10.(iii) and (3.4) now imply that the Ω3–component of d(XyReΩ) is 6 − 1α(dJX)+3X ∧ω 2 (cid:16) (cid:17) as claimed. On the other hand, by (3.3) and the fact that ω2 is closed d(XyReΩ)∧ReΩ = dX ∧ω2, d(XyReΩ)∧ImΩ = dJX ∧ω2. (iv) now follows from (3.5) and Lemma 2.10.(iv). In order to prove (v) observe that XyReΩ = JXyImΩ and therefore by (3.1), d∗(XyReΩ) = ∗d(JX ∧ReΩ)= ∗(dJX ∧ReΩ). Lemma (2.10).(iii) concludes the proof. (cid:3) We actually need a bit more: the Ω3 component of d(XyReΩ) is equal to the Ω3 component 12 12 of d∗(X ∧ReΩ). In particular, if d(XyReΩ)∈ Ω3 then d(XyReΩ)= 0. 12 Lemma 3.7. For every vector field X ∈X(M) we have d(XyReΩ)= (Jα(dX)+2X)∧ω−d∗XReΩ−d∗(JX)ImΩ+d∗(X ∧ReΩ). Proof. If ρ= L ReΩ then the linearisation of Hitchin’s duality map ρˆmust be X ρˆ= L ImΩ = d(XyImΩ)−4JX ∧ω = −d(JXyReΩ)−4JX ∧ω. X On the other hand, by Proposition 3.6.(iv) ρ = d(XyReΩ)= − 1α(dJX)+3X ∧ω− 1d∗XReΩ− 1d∗(JX)ImΩ+ρ 2 2 2 0 (cid:16) (cid:17) and Proposition 2.12 now implies ρˆ= − 1Jα(dJX)+3JX ∧ω+ 1d∗(JX)ReΩ− 1d∗XImΩ−∗ρ 2 2 2 0 = −(Jα(cid:16)(dJX)+6JX)∧ω(cid:17)+d∗(JX)ReΩ−d∗XImΩ−∗d(XyReΩ). Comparing the two expressions for ρˆwe conclude that d(JXyReΩ) =(Jα(dJX)+2JX)∧ω−d∗(JX)ReΩ+d∗XImΩ+∗d(XyReΩ). Up to changing X into JX, the Lemma is now proved since by (3.1) d∗(JX ∧ReΩ)= ∗d(XyReΩ). (cid:3) Remark 3.8. Equating the two different ways of writing ρˆalso yields the identity (3.9) α(dX) = Jα(dJX)+4JX of [16, Lemma 3.2]. Note also that integrating by parts the identity of Lemma 3.7 with a compactly supported X yields ∗ (3.10) α(d ρ) = −Jα(∗dρ) for every ρ ∈ Ω3 (M). In particular, if ρ∈ Ω3 is coclosed then dρ ∈ Ω4. Indeed, since ρ∧ω = 0= 12 12 8 ρ∧ReΩ by Lemma 2.11 we also know that dρ has not component in Ω4. 1 DEFORMATION THEORY OF NEARLY KÄHLER MANIFOLDS 9 3.2. The Dirac operator on nearly Kähler manifolds. Every 6–manifold M with an SU(3) structure is spin and it is endowed with a unitspinorψ. As an SU(3) representation the real spinor bundle S/(M) is isomorphic to Λ0⊕Λ0⊕Λ1, where the isomorphism is (f,g,X) 7−→ fψ+gVol·ψ+X ·ψ. For comparison with the Dirac-type operator we will define in the next subsection, we want now to deriveaformulafortheDiracoperatorD/ onanearlyKählermanifoldintermsofthisisomorphism. Recall that onanearly Kähler manifold wecan assumethedefiningunitspinorψ satisfies (2.16). In particular, D/ψ = −3ψ and D/(Vol·ψ) = 3(Vol·ψ). Thus D/ (fψ+gVol·ψ) = −3fψ+3gVol·ψ+(∇f −J∇g)·ψ, since the almost complex structure J satisfies (3.11) JX ·ψ = Vol·X ·ψ = −X ·Vol·ψ. On the other hand, 6 D/(X ·ψ) = e ·∇ X ·ψ−X ·ψ−X ·D/ψ =dX ·ψ+(d∗X)ψ+2X ·ψ. i ei i=1 X NowdecomposedX intotypes:dX = 1d∗(JX)ω+1α(dX)yReΩ+π (dX).Wehavetodetermine 3 2 8 the action of 2–forms on ψ. Lemma 3.12. For any 2–form η =λω+YyReΩ+η with η ∈ Ω2 we have 0 0 8 η·ψ = 3λVol·ψ+2JY ·ψ. Proof. Forms of type Ω2 are exactly those that act trivially on ψ . On the other hand, writing 8 0 ω = 3 e ∧Je for an SU(3)–adapted orthonormal co-frame {e ,Je ,e ,Je ,e ,Je }, i=1 i i 1 1 2 2 3 3 P 3 3 3 ω·ψ = (e ∧Je )·ψ = e ·Je ·ψ = − e ·e ·Vol·ψ = 3Vol·ψ. i i i i i i i=1 i=1 i=1 X X X Here we used [3, Equation (1.3)], the fact that e is orthogonal to Je and (3.11). i i In order to calculate the action of Ω2, observe that the intrinsic torsion of a nearly Kähler 6 structure is −ImΩ [16, p. 3] and that this acts as 4 on ψ and annihilates spinors of the form X·ψ, cf. for example [5, Lemma 2]. Thus, using [3, Equations (1.3) and (1.4)], (YyReΩ)·ψ = (JYyImΩ)·ψ = −1(JY ·ImΩ+ImΩ·JY)·ψ = 2JY ·ψ. (cid:3) 2 Proposition 3.13. For every f,g ∈ C∞(M) and X ∈ X(M) we have D/(fψ+gVol·ψ+X·ψ) = (d∗X−3f)ψ+(d∗JX +3g)Vol·ψ+(∇f −J∇g−α(dJX)−2X)·ψ. Proof. Use Lemma 3.12 and (3.9) to rewrite Jα(dX) as −α(dJX)−4X. (cid:3) 3.3. A Dirac-type operator. Consider now the first order operator D: Ω1⊕Ω0⊕Ω0 → Ω1⊕Ω0⊕Ω0 defined by (3.14) D(X,f,g) = 1α(dJX)+3X +df +Jdg,d∗X +6f,d∗(JX)−6g . 2 (cid:16) (cid:17) Proposition 3.15. D is a self-adjoint elliptic operator. Proof. Proposition 3.13 shows that, after choosing appropriate isomorphisms between S/(M) and Λ0⊕Λ0⊕Λ1 we can identify D with D/ up to a self-adjoint zeroth order term. (cid:3) 10 L.FOSCOLO Remark 3.16. For every s ∈ R one can define a Dirac operator D associated with the connection s ∇s = ∇+sT, where ∇ is the Levi-Civita connection and T = −ImΩ is the intrinsic torsion (when 2 s = 1 this is theso-called canonical Hermitian connection). None of these Dirac operators coincides with the operator D defined in (3.14). Our interest in the operator D arises from the following identities. Consider the operator (3.17) D+: Ω2 ⊕Ω4 −→ Ω3 1⊕6 1 1⊕1⊕6 defined by (fω+XyReΩ,12gω2) 7−→ π1⊕1⊕6 d(fω−XyReΩ)+ 12d∗(gω2) . Since by Proposition 3.6 the image of (fω+XyR(cid:16)eΩ,1gω2) is (cid:17) 2 1α(dJX)+df +Jdg+3X ∧ω+ 1 (d∗X +6f)ReΩ+ 1(d∗JX −6g)ImΩ, 2 2 2 D+ coincid(cid:16)es with D after choosing a(cid:17)ppropriate identifications of Ω2 ⊕ Ω4 and Ω3 with 1⊕6 1 1⊕1⊕6 Ω0⊕Ω0⊕Ω1. Similarly, the operator (3.18) D−: Ω3 −→ Ω4 ⊕Ω2 1⊕1⊕6 1⊕6 1 defined by ∗ σ = JX ∧ω−gReΩ+fImΩ 7−→ (π1⊕6dσ,π1d σ) coincides with D after identifying Ω4 ⊕Ω2 with Ω0⊕Ω0⊕Ω1 by 1⊕6 1 (f,g,X) 7−→ (XyReΩ)∧ω− 1fω2,2gω . 3 3 Much of what follows relies on the ma(cid:16)pping properties of the opera(cid:17)tor D (and therefore of D±). Proposition 3.19. Let (M,g,ω,Ω) be a complete nearly Kähler 6–manifold not isometric to the round 6–sphere. Then the kernel (and cokernel, since D is self-adjont) of D consists of Killing fields that preserve the whole SU(3) structure. Proof. Suppose that (X,f,g) lies in the kernel of D. Then (a) 1α(dJX)+df +Jdg+3X = 0; 2 (b) d∗X +6f = 0; (c) d∗(JX)−6g = 0. We apply d∗ ◦ J to the identity (a): using (ii) and (v) in Proposition 3.6 and (c) we obtain d∗dg+18g = 0 (recall Remark 2.5!) and therefore g = 0. Hence (a’) 1α(dJX)+df +3X =0; 2 (b’) d∗X +6f = 0; (c’) d∗(JX) = 0. Now, (a’) and (b’) together with (3.5) imply dJX = −1(d∗X)ω+ 1α(dJX)yReΩ+π (dJX) = 2fω−(df +3X)yReΩ+π (dJX). 3 2 8 8 Using Proposition 3.6 we differentiate this identity and take the inner product with ReΩ: 0 = 1hd2(JX),ReΩi = 6f + 1d∗(df +3X) = 1(d∗df −6f). 4 2 2 By Obata’s Theorem f = 0 under the assumptions of the Proposition. It remains to show that a vector field X such that dJX = −3XyReΩ+π (dJX) and d∗(JX) = 8 0 = d∗X must preserve the SU(3) structure. Since d∗X = 0 and thus dJX ∧ω2 = 0, by Lemma 2.10.(i) we have 18kXk2 −kπ (dJX)k2 = dJX ∧dJX ∧ω = 3 JX ∧dJX ∧ReΩ = 18kXk2 . L2 8 L2 ˆ ˆ L2 M M Thus 0 = dJX + 3XyReΩ = L ω. Moreover, by Proposition 3.6.(iv) d(XyReΩ) ∈ Ω2 and X 12 therefore the formula of Lemma 3.7 implies that L ReΩ = d(XyReΩ)= 0. (cid:3) X