Quaternionic K¨ ahler Metrics

connectionA. IfM is a hyperK¨ahler 4-manifold, the symplectic forms span Λ²₊T^∗M and we obtain a formal hyperK¨ahler quotient of A by G

µ⁻¹(0)

G = {A :F_A ∈Ω²₊(g_P)}

G ,

which is the moduli space of self-dual solutions of the Yang-Mills equations. When M is R⁴, this space may also be obtained as a finite-dimensional hyperK¨ahler quo- tient via the adhm construction (Atiyah et al., 1978b) which describes the moduli space in terms of matrix algebras. The remark at the end of Section 3.3 shows that these moduli spaces are associated to quaternionic K¨ahler manifolds.

3.5 Quaternionic K¨ahler Metrics 64 i) there is a finite subgroup Γ of Sp(1) such that Sp(1)/Γ acts freely

on N,

ii) Sp(1)induces a transitive action on the2-sphere of complex structures on N, and

iii) if XI is the vector field generated by the circle subgroup of Sp(1) preservingI, then the (real) linear span ofIX_I inT N is independent of the choice of complex structure I.

Choose a subgroup U(1) 6 Sp(1) preserving a complex structure I. Let µ: N → R be a moment map for this U(1) with respect to the K¨ahler structure defined byI, thenµ⁻¹(x)is Sp(1)-invariant andµ⁻¹(x)/Sp(1)is a quaternionic K¨ahler manifold.

Proof. Fix a complex structure I preserved by a subgroup U(1)6Sp(1) and let ω_I denote the K¨ahler form associated to I. Let X be the vector field generated by the U(1)-action and let Y be a vector field generated by the Sp(1)-action. Then Y is orthogonal to IX, since Y arises from a circle group preserving a complex structure A and Y is orthogonal toAY =IX. Now,

Yydµ=Yy(Xyω_I) =−g(Y, IX) = 0,

so µ is Sp(1)-invariant.

We now give two different proofs that the quotient µ⁻¹(x)/Sp(1) is quater- nionic K¨ahler. The first shows that we have a closed quaternionic 4-form on the quotient and then uses Theorem 1.2.2 to deduce that it is quaternionic K¨ahler. This is only valid if the quotient manifold has dimension at least 12. The second proof applies in all dimensions and is more in the spirit of Galicki & Lawson’s proof (1988) of the quaternionic K¨ahler quotient construction.

LetJ andK be complex structures such thatIJ =K =−JI and letωJ,ωK be the corresponding K¨ahler forms. Define a 4-form by Ω =ω_I∧ω_I+ω_J∧ω_J+ω_K∧ω_K. Leti: µ⁻¹(x)֒→N andπ: µ⁻¹(x)→µ⁻¹(x)/Sp(1) be the inclusion and projection respectively. If n is a zero of µ, then we can write T_nN = Vn ⊕ Hn where V is the quaternionic span of X and H is the orthogonal complement. Note that this splitting is quaternionic. The hypotheses of the theorem imply that V contains all vectors tangent to the Sp(1)-action and that H is an Sp(1)-invariant distribution of horizontals for the projection π. The restriction i^∗Ω of Ω vanishes on V and is Sp(1)-invariant, so i^∗Ω is the pull-back of a 4-form Ω^′ on the quotient. Since H is quaternionic and π^∗Ω^′ is just the restriction of Ω on H, we see that Ω^′ is of the correct algebraic type to define a quaternionic structure on the quotient. Now i^∗Ω =π^∗Ω^′ and π^∗ is injective, so dΩ = 0 implies that Ω^′ is closed. This concludes the proof if the quotient is at least 12-dimensional.

For the second proof we need to work with the covariant derivative ∇^′ on the quotient. As above, for n ∈ µ⁻¹(x) split TnN as Vn ⊕ Hn, write i and π for the inclusion and projection maps, and let ·^H denote the horizontal component of a tangent vector. Then if Z ∈ Hn is the pull-back of a tangent vector Z on the quotient, we have

∇^′Z =π_∗((∇Z)^H),

where ∇ is the Riemannian connection on N. Note that ifω_A is one of the K¨ahler forms on N, then if Z ∈ H and Y is any tangent vector, we have ω_A(Y, Z) = ω_A(Y^H, Z). Choosing a local section s of π we obtain a local 2-form s^∗ω_A on the quotient and there we obtain a rank 3 vector bundleG generated by all such local 2- forms. Except in 4-dimensions, it is sufficient to show that this bundle is preserved by∇^′. If ω_A is such a local 2-form, thenπ^∗ω_A= (f_Iω_I +f_Jω_J+f_Kω_K)^H for some

3.5 Quaternionic K¨ahler Metrics 66 functions fI, fJ, fK. Now

π^∗∇^′ωA = (∇((π^∗ωA)^H))^H

= (∇(fIωI +fJωJ +fKωK)^H)^H

= (df_I ⊗ω_I +df_J ⊗ω_J +df_K ⊗ω_K)^H,

since by the remark above, (∇(ω_I)^H)^H = (∇ω_I)^H = 0. Thus ∇^′ω_A is in G, as required.

In 4-dimensions we need to calculate the curvature. Ifs is a local section, then s^∗ωI defines a complex structure if and only if it is a holomorphic section with respect to I. As in the integrability proof in the previous chapter, this implies that the curvature tensor lies in the complement of Λ^0,2 ⊗Λ^0,2. This holds for each complex structure on N and forces the curvature to be self-dual. The Einstein condition follows from an immersion computation and the formulae of O’Neill (1966) applied to the Riemannian submersion µ⁻¹(x)→µ⁻¹(x)/Sp(1).

The above theorem generalises to the pseudo-Riemannian category in the same way as we remarked that the hyperK¨aher quotient construction could be generalised.

The hypotheses on the group action in the above theorem, imply that the action is essentially determined by aU(1)-action and the quaternionic structure ofN. This is a quaternionic analogue of having a complexified U(1)-action, which is what is used in the ordinary K¨ahler quotient construction. Note that an Sp(1)-action also determines much of the hyperK¨ahler structure of N.

Lemma 3.5.2. Let (N, g) be a Riemannian manifold with symplectic formω. SupposeN admits an isometric SU(2)-action such that the distri- bution A spanned by SU(2)·I is 3-dimensional and the orthogonal com- plement I^⊥ in A is preserved by the almost complex structure I defined by ω andg. ThenN is locally hyperK¨ahler.

Proof. Locally there is a g∈ SU(2) such that g^∗ω is orthogonal to ω. Define ωI

to beω and putω_J =g^∗ω. DefineJ byg(X, JY) =ω_J(X, Y) for tangent vectorsX andY. Then, theI-invariance of the decomposition ofAimplies thatIJ ∈SU(2)·I and hence that (IJ)² = −1, so IJ = −JI. Define K to be IJ and define ω_K in the same way as ω_J. Then ω_I, ω_J and ω_K are closed 2-forms and so define a local

hyperK¨ahler structure.

The product of two quaternionic K¨ahler manifolds need not be quaternionic, but the above theorem suggests a type of quaternionic join construction. Let M₁, M₂ be quaternionic K¨ahler manifolds with positive scalar curvature. Con- sider the bundles U(M₁), U(M₂) with their hyperK¨ahler metrics. The product U(M₁)× U(M₂) is hyperK¨ahler and H^∗ acts diagonally. The Sp(1)-action so de- fined satisfies the hypotheses of the theorem and we obtain a quaternionic K¨ahler manifold J(M₁, M₂) of dimension dimM₁ + dimM₂ + 4. Topologically this is the Z₂-quotient of the quaternionic projectivisation of the bundle which is locally π₁^∗H(M1)⊕π₂^∗H(M2)→M1×M2, whereπi is projection onto theith-factor. Thus if M₂ is a point ∗, J(M₁,∗) is topologically U(M₁).

Proposition 3.5.3. Let M be a quaternionic K¨ahler manifold with pos- itive scalar curvature. The quaternionic K¨ahler metric onJ(M,∗) agrees with the metricg₁ constructed on U(M) in Chapter 2.

Proof. First note that the join of two points J(∗,∗) is the Z₂-quotient of HP(1) with two points removed with its standard symmetric metric g^HP(1). In inhomoge- neous coordinates,

g^HP(1) = Re

d¯z ⊗dz

1 +kzk² − zdz¯ ⊗d¯zz (1 +kzk²)²

= Re

d¯z⊗dz (1 +kzk²)²

.