Commun. Math. Phys.

125,637-642 (1989)

Communications in

Mathematical Physics

Springer-Verlag 1989

Complete Ricci-Flat Kahler Manifolds of Infinite Topological Type

Michael T.Anderson 1 '*, Peter B.Kronheimer2'**, and Claude LeBrun 1 '*
Department of Mathematics, SUNY at Stony Brook, Stony Brook, NY 11790, USA School of Mathematics, Institute for Advanced Study, Princeton, NJ 08540, USA and Merton College, Oxford, 0X1 4JD, United Kingdom
2 1

Abstract. We display an infinite dimensional family of complete Ricci-flat Kahler manifolds of complex dimension 2, for which the second homology is infinitely generated. These are obtained from the Gibbons-Hawking Ansatz [2] by using infinitely many, sparsely distributed centers.
Introduction

In [2], Gibbons and Hawking construct families of complete Ricci-flat Kahler metrics on a class of non-compact 4-manifolds Nk. The metrics are asymptotically locally Euclidean in the sense that dNk& S3/Zk, and the metrics approach, at infinity, the locally Euclidean metric on the cone C(S3/k). Another description of these metrics was given by Hitchin [3]. Further examples, with boundary a spherical space form S3/, a SE/(2), and a characterization of these metrics (Torelli theorem) among asymptotically locally Euclidean metrics were obtained by Kronheimer [5,6]. In this paper, we show that one may also obtain complete Ricci-flat Kahler metrics corresponding to the case "k= GO" of the Gibbons-Hawking metrics. These metrics are no longer asymptotically locally Euclidean, or of finite action, nd and are carried by a 4-manifold whose 2 homology is infinitely generated. It is only recently (7) that examples of complete metrics of non-negative Ricci curvature have been exhibited on manifolds of infinite topological type. The example shows that a complex 2-manifold supporting a complete Ricciflat Kahler metric need not be the complement of a divisor in a compact complex surface since the homology of such a complement is certainly finitely generated. This indicates that a conjecture of Yau [8,9] concerning the existence of such compactifications is not true without some strengthening of the hypothesis. These metrics also provide the first example for which the moduli space of complete Ricci-flat metrics on a given manifold is infinite dimensional.
* Partially supported by N.S.F. grants DMS 87-01137 and DMS 87-04401 ** Partially supported by N.S.F. grant DMS 86-10730

638

M.T. Anderson, P.B. Kronheimer, and C. LeBrun

1. Construction of the Manifold

We begin by considering any divergent sequence of distinct points p} e R 3 , y e N . We will construct a 4-manifold M and a smooth map : M -* R 3 such that ~1 (p^ is a point for ally, but " 1 ^ ) ^ ^ 1 for pelR3 {Pj}. To begin, we let 0 M 0 - ^ R 3 {pj} be the principal S1 bundle whose Chern class is 1 when restricted to a small sphere around anyp,-; here "small" means of radius less than
Pir PiW. S i n c e

7=1

is the free abelian group generated by the homology classes of these small spheres, this uniquely determines the Chern class in
.7=1

and thus determines a unique principal S bundle. Thus TIQ1 (Br (pj)) is diffeomorphic to a punctured 4-ball B} {0} cz R 4 in a manner such that the S1 action becomes the action of S1 a C on C 2 = R 4 by scalar multiplication. We then M=Mou (J Sj := M0U
J = l

U
.7=1

$j/~,

where the equivalence relation ~ identifies Bj {0} with 0 x (^r (?_/)) The map 0 : M o -^R 3 clearly extends to a smooth map : M->R 3 . Note that there is an S 1 action on Mand is just the projection to the orbit space, with {pj} corresponding to the fixed points of the action. To understand better the topology of M, consider the case in which the points Pj in the description above are given by Pj = (x7 ,0,0), with Xj<xj+1 and let Dj = ~1([pj9pj+1]) be the inverse image of the line segment [Pj,Pj+]^L3. Each Dj is a smoothly embedded 2-sphere with self-intersection 2, meeting Dj+ transversely at the point n"1 (pj+1). Clearly, the manifold M is diffeomorphic to the open subset N a M consisting of the tubular neighborhood of these spheres. It follows that M is simply connected and
L

HJM,Z) = 0 otherwise.

This description can be summarized by saying t h a t M is the result of plumbing an infinite family of 2-spheres according to the " C a r t a n m a t r i x " 2

1 -2 1

1 -2 1

Ricci-Flal Kahler Manifolds

639

Note that the Gibbons-Hawking metrics with k centers correspond to plumbing a collection of (k 1) 2-spheres according to the Cartan matrix of Ak, cf. [1, 3].

2. The Gibbons-Hawking Metric

We now restrict somewhat the above choice of the sequence {Pj}? in IRA Namely, we impose the extra condition that, for some point /? o eR 3 we have

for example, we might take/?7 = (/2,0,0) and let/?0 = (0,0,0). It then follows that V: R 3 - {pj} ->R defined by

is a smooth function on IR3 {pj}. Clearly, Fis a solution of the Laplace equation

where * is the Hodge * operator on R 3 . Further, it is easily verified that the cohomology class of the closed 2-form -= *dV represents the Chern class of the In principal S1 bundle 0 : M 0 -R 3 {pj} in deRham cohomology. There is therefore a connection on 0: M 0 -R 3 {pj} with curvature *dV. Let e1 (M o ) be the connection 1-form for such a connection, so that The form is then unique up to gauge transformations, since R 3 {? ,} is simply / _. connected. The Gibbon-Hawking metric on Mo is given by
g = 77 co co + V% ds2 ,

where ds2 is the Euclidean metric on R 3 . It has anti-self dual curvature tensor, as follows from d = $ (*dV), see for example [4]. In particular g is Ricci-flat. Since M o is simply connected, it follows that Mo is hyperkahler, (cf. [4]), i.e. there is an entire 2-sphere's worth of complex structures for which g is a Kahler metric. To display these parallel complex structures explicitly, let e1,e2, e3 be any oriented orthonormal basis for R 3 . Consider these as constant vector fields on R 3 and let e1 ,e2, e3 be their horizontal lifts to Mo via the connection . Further, let X denote the generator of the S action on Mo. Then V1/2X, V-1/2e^ V~1/2
2

640

M.T. Anderson, P.B. Kronheimer, and C. LeBrun

is an orthonormal frame for Mo. Relative to this frame, the matrix

0 1 0 0

-1 0 0 0

0 0 0 1

0 0
j

defines an almost complex structure, depending only on the choice of ex, which one may verify to be parallel, and hence integrable. The Gibbons-Hawking metric now continues smoothly across the isolated points
r

Indeed, near pj9

an

we have

V= y- + / = Vo +/, where

(p)

1 1

i i j) ~ {Pj})) with d0 = (*dV0)9 then it is easily seen that the metric {} ih d) i i i h h

V0(ds2)

\\pPj\\> d where / is smooth. If 0 is the connection form on j

extends smoothly over $ M. In fact, g0 is just the flat metric defined near a s o n e sees o (Pj\ by performing the coordinate change r-> ]/2r. Clearly, the metrics g and g0 differ by a smooth bilinear form, depending on / only, so that g extends smoothly to M. It follows that the curvature tensor in again anti-self dual, and, since M is simply connected, this makes M hyperkahler. As a consequence, any of the parallel complex structures on M o extends as a parallel complex structure to M. Choosing one makes (M,g) a Ricci-flat Kahler surface. As a particular case, suppose again that Pj = (^.,0,0), with Xj<xj+ and /\Xj\ < oo. If e points along the x-axis, then the 2-sphere Dj described in Sect. 1 is a holomorphic curve with respect to the complex structure defined above. If, on the other hand, we consider the complex structure corresponding to any other 3 direction in R , then M contains no holomorphic curves: for example, e points along the z-axis, then M becomes biholomorphically equivalent to the hypersur3 face in (C defined by the equation

Briefly, to see this, note that the projection, , of M onto the (e2,e3) plane, thought of as (C, is holomorphic. This defines the coordinate 3 above. The fibre 1 1 ~ (C3) is generically one orbit of the <C* action, the complexification of the S 1 action on M defined in Sect. 1. Note that (C* - {x C? = 1} < & - The only = exception is where 3 is the image of one of the p^ in which case ~ (3) is the curve (i 2 = 0. This description is precisely analogous to the description [3] of the complex manifolds arising from the Gibbons-Hawking ansatz in the case of finitely many centers. In view of the classification scheme of [6], one might expect a similar limit for the family of gravitational instantons corresponding to the Cartan matrices Dk as k -> oo.

Ricci-Flat Kahler Manifolds

641

3. Completeness Let z n e M b e a Cauchy sequence with respect to g; let yn = (zn) denote the projection of the sequence to R 3 . We claim that {zn} converges. If not, we have yn /?! for all but finitely many n, so without loss of generality, yn p1 for all n. Let denote distance in R 3 {px} with respect to the metric (ds2)/2r, where r(p) = || p /?! || let denote distance in M with respect to g. Then for any a,beM we have (a,b)>((),(b)). Indeed, it suffices to observe that for any curve in Mo, the length of with respect to Vds2 is less than that of with respect to g, since g was constructed so as to make (M 0 ,g)-(R 3 - {pj}, Vds2) a Riemannian submersion. But since V> l/2r, is even shorter with respect to (ds2)/2r. Since S(a,b) is just the infimum of the lengths of curves joining a and b in M o , the inequality follows. Thus yn is a Cauchy sequence with respect to ds2/2r. We claim that \\p yn || is therefore bounded. Indeed, we know that for some C, (y,yn)<C for all ft, since the sequence is Cauchy; but for any curve : [, b]->R3 {Pi} we ds2 have that the length of with respect to ^ satisfies

^ J ^
() ]/2r so that (y, yn) ^ j/2 ||/r(j) j/r (jv) |.

]/2 \]/r) -

/c

\2

Hence r (yn) < = + }/r{y)\ =R for all n, and yn is a bounded sequence. It follows that for some R the sequence {zn} is contained in " 1 (BR{p)). Since this is compact, it follows that {zn} converges. To summarize, we have proved Theorrem. (M, g) is a complete, hyperkahler 4-manifold with infinitely generated homology group H2. We note finally that we have produced an infinite-dimensional family of such metrics on M. Indeed, as in the finite case [3], the configuration of points {pj} can be uniquely recovered from the metric g, to within an isometry of R 3 . One way to prove this is to observe first that the natural isometric circle action on M is uniquely determined as being the only circle action to preserve all the complex structures. (Hyperkahler 4-manifolds with more than one such circle action can be classified, and ours is not on the list.) The projection : M-R 3 is then the momentum mapping for this action, in the sense of [4], and the configuration {pj} is the image of the fixed point set.

642

M.T. Anderson, P.B. Kronheimer, and C. LeBrun

References
1. Barth, W., Peters, C , Van de Ven, A.: Compact complex surfaces. Ergeb. Math., 3. Folge, Bd 4. Berlin, Heidelberg, New York: Springer 1984 2. Gibbons, G., Hawking, S.: Gravitational multi-instantons. Phys. Lett. 78B, 430-432 (1978) 3. Hitchin, N.: Polygons and gravitons. Math. Proc. Camb. Phil. Soc. 85, 465-476 (1979) 4. Hitchin, N., Karlhede, A., Lindstrom, U., Rocek, M.: Hyperkahler metrics and supersymmetry. Commun. Math. Phys. 108, 535-589 (1987) 5. Kronheimer, P.: Instantons gravitationelles et singularites de Klein. C. R. Acad. Sci. Paris 303, Ser. I 53-55 (1986) 6. Kronheimer, P.: A Torelli type theorem for gravitational instantons. J. Diff. Geo. 29, 3 (1989) 7. Sha, J.-P., Yang, D.G.: Metrics of positive Ricci curvature on connected sums of Sn x Sm (preprint) 8. Yau, S.-T.: Problem section, seminar on differential geometry. Ann. Math. Studies, Vol. 102. Princeton: Princeton University Press 1982 9. Yau, S.-T.: Nonlinear analysis in geometry. L'Enseignement Math. 33, 109-156 (1987) Communicated by S.-T. Yau Received January 5, 1989