betaFeedback

WorksMathematics

Spherical space forms in K3: an instanton parity argument

Abstract

Let RR be a connected subdiagram of the Dynkin diagram E8E_8, and let YRY_R be the spherical three-manifold which bounds the corresponding plumbing of (−2)(-2)-spheres. We show that if a negative-definite four-manifold NN with H1(N)=0H_1(N)=0 and oriented boundary YRY_R is embedded in the K3 surface, then the intersection form of NN is the negative-definite root lattice −R-R. As a consequence, the traces of −2-2 and −4-4 surgery on the left-handed trefoil do not embed in K3. Heegaard Floer correction terms and the theory of lattices reduce the theorem to the exclusion of one smaller lattice for each of R=D5R=D_5, D6D_6, D7D_7 and E7E_7. In these cases, stretching the K3 surface along YRY_R exhibits a zero-dimensional moduli space of instantons as the zero set of a section of a real two-plane bundle over a compact surface. The quaternionic nature of the linearized anti-self-duality equations on K3 shows that the Euler number of this bundle is even, while the corresponding Donaldson invariant is odd.

Foreword

With the exception of this foreword, the present paper is entirely AI-generated, in response to prompts provided by the human authors. The mathematics was produced by GPT-6 Astra, and the paper was redrafted by Claude Fable 5.2. The main result is Theorem 1.1. A particular case states that the 4-manifold NN obtained by adding a 2-handle to the 4-ball along a left-handed trefoil with framing −2-2 cannot be smoothly embedded in a K3 surface. (See Corollary 1.2.)

These results arose from prompts motivated by a question which the authors learned of from Anubhav Mukherjee. Let SS be the standard 3-sphere in the neck of the connected sum K3#K3K3\#K3. Does the connected sum contain any sphere different from SS? Here one can interpret different as up to isotopy; or one can ask whether ±[S]\pm[S] is the only non-zero class in π3\pi_3 that is represented by an embedding. In the latter version, the question is closely related to the question of whether a diffeomorphism of K3#K3K3\#K3 can “mix” the homology of the two summands. The clearest strategy to construct such a mixing diffeomorphism starts with a hypothetical embedded 2-sphere Σ\Sigma in K3#K3K3\#K3 with Σ⋅Σ=−2\Sigma\cdot\Sigma=-2 such that the class [Σ]∈H2(K3)⊕H2(K3)[\Sigma]\in H_2(K3)\oplus H_2(K3) has non-zero projection onto both sides. Then reflection in [Σ][\Sigma] arises from a diffeomorphism of K3#K3K3\#K3 which does not preserve the direct sum decomposition. (Indeed, if [Σ][\Sigma] is the sum of two such primitive classes, then it turns out that this would imply that the diffeomorphism group of K3#K3K3\#K3 acts by the full isometry group of the homology lattice.) To try and construct such a 2-sphere, one might start with a cuspidal elliptic fiber on one side, and seek on the other side an embedded disk with relative self-intersection number −2-2 and boundary a trefoil in the 3-sphere SS. Tubing the trefoil in SS to the cusp gives the sought-after 2-sphere. But finding the disk in punctured K3K3 is equivalent to embedding the 4-dimensional handlebody NN mentioned above, which Corollary 1.2 says cannot be done. Of course, only the simplest construction of such a sphere is ruled out by this argument, because the hypothetical sphere Σ\Sigma might meet SS in an arbitrary knot or link instead of the trefoil.

ChatGPT proved the corollary first, for the case of the −2-2 framed left-handed trefoil. It began by giving an argument conditional on the superconformal simple type conjecture [12, 7]. If NN embeds in K3K3, then the manifold obtained by replacing this piece with an E7E_7 plumbing is a manifold whose Seiberg-Witten invariants can be computed sufficiently to obtain a contradiction to that conjecture. The AI then found an unconditional proof using Donaldson invariants, which is the one presented here.

In outline, here is the proof by contradiction as it applies to a hypothetical embedding of N=X−2(T−)N=X_{-2}(T_-) in K3K3. With a suitable SO⁡(3)\operatorname{SO}(3) bundle on K3K3, one obtains a zero-dimensional ASD moduli space, and the Donaldson invariant is known to be ±1\pm1. On the other hand, stretching the metric along ∂N\partial N leads to a gluing description which presents the moduli space as the zero set of a section of an obstruction bundle L→ML\to M, where LL has real rank 2 and MM is a compact, oriented 2-manifold. This describes the Donaldson invariant as the Euler class ⟨e(L),[M]⟩\langle e(L),[M]\rangle. However, [L]−[TM][L]-[TM] has an interpretation as the element of KO(M)KO(M) arising from the index of the ASD deformation complex. The quaternionic structure of the tangent bundle of K3K3 implies that this element is even, and it follows that the Donaldson invariant of K3K3, computed as ⟨e(L),[M]⟩\langle e(L),[M]\rangle, is even, which contradicts the known result.

The justification of the required gluing result in section 5 of the paper is verbose. The AI initially found no suitable models for the required analytic steps in the literature. The human authors pointed to the cited papers of Taubes, which provided both the overall strategy for the gluing theorem and the analytic details. With this background, GPT-6 wrote a gluing argument which was later condensed by Fable.

After completing the proof of the corollary, the generalization given in Theorem 1.1 was suggested by the authors, as something that might follow from similar arguments. As noted in the paper, the instanton parity argument is needed only in the cases D5D_5, D6D_6, D7D_7 and E7E_7, to rule out the lattices ⟨−4⟩\langle-4\rangle, −(A1⊕A1)-(A_1\oplus A_1), −A3-A_3 and −A1-A_1 respectively.

The above corollary and the parity argument in its proof are both new to the authors. Distributing the preprint in this form, the authors would be glad to hear from others if these results are known.

Acknowledgements.

The work of the first author was supported by NSF grant DMS-2304877. Both authors were supported by Simons Foundation Awards #994330 and #14760-09, Simons Collaboration on New Structures in Low-Dimensional Topology.

Introduction

Let RR be a Dynkin diagram of type AA, DD or EE. We use the same letter for the corresponding root lattice, which we take to be positive definite. Let PRP_R be the four-manifold obtained by plumbing disk bundles of Euler number −2-2 over the two-sphere according to the diagram RR, so that the intersection form of PRP_R is −R-R, and let YR=∂PRY_R = \partial P_R be its oriented boundary. Thus YRY_R is the quotient of S3S^3 by a finite subgroup of SU(2)SU(2). This paper is concerned with embeddings of the manifolds YRY_R in the K3 surface. Our main result is the following.

Theorem 1.1. Let RR be a connected subdiagram of the Dynkin diagram E8E_8, and let NN be a compact, connected, codimension-zero submanifold of the K3 surface whose oriented boundary is YRY_R. Suppose that NN is negative definite and that H1(N;Z)=0H_1(N;\mathbb{Z}) = 0. Then the intersection form of NN is isomorphic to −R-R.

The diagrams RR to which the theorem applies are AnA_n for 1≤n≤71 \le n \le7, DnD_n for 4≤n≤74 \le n \le7, and E6E_6, E7E_7 and E8E_8. With the hypotheses of the theorem, the group H2(N;Z)H_2(N;\mathbb{Z}) is free and H3(N;Z)H_3(N;\mathbb{Z}) is zero (Proposition 7.1), so NN has the same integral homology as the plumbing PRP_R.

In three cases, the conditions which the theorem imposes on NN do not need to be assumed. If RR is E7E_7, D5D_5 or D7D_7, and YRY_R is embedded in the K3 surface in any manner, then it divides the K3 surface into two pieces, and one of the two is negative definite, has oriented boundary YRY_R, and has vanishing first homology. Theorem 1.1 applies to that piece, whose intersection form is therefore −R-R. See Theorems 2.1 and 7.2.

For a knot K⊂S3K \subset S^3 and an integer mm, let Xm(K)X_m(K) denote the trace of mm-surgery on KK: the four-manifold obtained by attaching a two-handle to the four-ball along KK with framing mm.

Corollary 1.2. Let KK be the left-handed trefoil. Then the trace X−2(K)X_{-2}(K) does not admit an orientation-preserving smooth embedding in the K3 surface. The same is true of X−4(K)X_{-4}(K).

Proof. As oriented manifolds, the boundary of X−2(K)X_{-2}(K) is YE7Y_{E_7} [1 Lemma 1.2], and the boundary of X−4(K)X_{-4}(K) is YD5Y_{D_5} [15]. The traces are simply connected, and their intersection forms are ⟨−2⟩\langle-2\rangle and ⟨−4⟩\langle-4\rangle respectively, rather than −E7-E_7 and −D5-D_5. □

Outline of the proof

The first part of the proof of Theorem 1.1 uses Heegaard Floer correction terms and the theory of lattices. The manifold NN is spin, so the inequality of Ozsváth and Szabó [18] bounds b2(N)b_2(N) in terms of the correction terms of YRY_R, and also determines b2(N)b_2(N) modulo 8. The classification of even definite lattices of small rank and determinant then shows that the intersection form of NN is −R-R when RR is D4D_4, E6E_6 or E8E_8. The same holds for R=AnR = A_n, by an argument which also uses Donaldson’s diagonalization theorem. In the four remaining cases, there is one possibility for the intersection form other than −R-R:

RE7D5D6D7other possibility for QN⟨−2⟩⟨−4⟩−(A1⊕A1)−A3(1)\begin{array}{c|cccc} R & E_7 & D_5 & D_6 & D_7 \\ \text{other possibility for }Q_N & \langle-2\rangle& \langle-4\rangle& -(A_1 \mathbin{\oplus} A_1) & -A_3 \tag*{(1)} \end{array}

To exclude these four possibilities we use instantons. Consider the case R=E7R = E_7, and suppose that a K3 surface XX is decomposed as N∪YCN \cup_Y C, where Y=YE7Y = Y_{E_7} and NN has intersection form ⟨−2⟩\langle-2\rangle. Let P→XP \to X be an SO(3)SO(3) bundle with p1(P)=−6p_1(P) = -6 whose second Stiefel–Whitney class is non-zero on the generator of H2(N)H_2(N). The moduli space of anti-self-dual connections in PP has formal dimension zero, and the corresponding Donaldson invariant, which counts the points of this moduli space with signs, is ±1\pm1. We examine the moduli space as the metric on XX is stretched along YY. On the manifold NN with a cylindrical end attached, the bundle PP carries a reducible anti-self-dual connection A0A_0 of action 1/81/8, asymptotic to a reducible flat connection α\alpha on YY. A dimension count shows that, when the neck is long, every anti-self-dual connection in PP is close to A0A_0 on NN, and on CC is close to a point of a moduli space MM of instantons asymptotic to α\alpha, where MM is a compact two-manifold. The connection A0A_0, however, belongs to a moduli space of formal dimension −3-3. Its stabilizer is a circle and, however the equations are perturbed, the cohomology group HA02H^2_{A_0} of its deformation complex is not zero: it is a copy of C\mathbb{C} on which the stabilizer acts with weight one. In this situation, gluing theory describes the moduli space on XX not as a copy of MM but as the zero set of a section of a real two-plane bundle E→ME \to M, the bundle of obstruction spaces. The Donaldson invariant is therefore equal, modulo 2, to the Euler number of EE.

The proof is completed by showing that this Euler number is even. The gluing construction provides a family of connections on XX parametrized by MM, and the index of the family of linearized operators dA∗⊕dA+d_A^* \oplus d_A^+ is the element [TM]−[E][TM]-[E] of KO(M)KO(M). For a hyperkähler metric on XX, the operator dA∗⊕dA+d_A^* \oplus d_A^+ can be identified with a Dirac operator coupled to the real vector bundle ad⁡P\operatorname{ad} P, and so commutes with a quaternionic structure. The index of any family of such operators therefore lies in the image of KSp→KOKSp \to KO, and in particular its second Stiefel–Whitney class is zero. It follows that w2(E)=w2(TM)w_2(E)=w_2(TM), so the Euler number of EE has the same parity as the Euler number of the surface MM, which is even.

The cases R=D5R=D_5, D6D_6 and D7D_7 are dealt with in the same way, and differ only in the numerical data.

The method of gluing in the presence of an obstruction goes back to Taubes [22] and Donaldson [3], and is a standard part of the theory [4, 5]. Obstructed gluing at a reducible connection, for cobordisms between rational homology spheres, is studied in [2 Section 5]. We have collected the analysis in § 6, and the reader who is willing to take Proposition 4.1 on trust can omit that section.

Organization

Organization. Sections 2–4 deal with the case R=E7R=E_7. § 3 contains the lattice-theoretic part of the argument, § 4 describes the moduli spaces on the two pieces, and § 5 contains the parity argument. The transversality and gluing results on which these depend are proved in § 6. The cases R=D5R=D_5, D6D_6 and D7D_7 are treated in § 7, and the remaining cases in § 8, where the proof of Theorem 1.1 is completed.

The case R=E7R=E_7: constraints on the lattice

In this section and the next two, YY denotes the manifold YE7Y_{E_7}. It is the quotient of S3S^3 by the binary octahedral group O∗O^*, a group of order 48 whose abelianization is Z/2\mathbb{Z}/2. For this manifold we prove a statement which is a little stronger than Theorem 1.1, in that the hypotheses on NN become conclusions.

Theorem 2.1. Suppose that Y=YE7Y=Y_{E_7} is smoothly embedded in a K3K3 surface XX. Then YY separates XX into two pieces,

X=W∪YC,X=W\cup_Y C,

where WW has oriented boundary YY and is negative definite. The groups H1(W;Z)H_1(W;\mathbb{Z}) and H1(C;Z)H_1(C;\mathbb{Z}) are zero, and the intersection form of WW is isomorphic to −E7-E_7. In particular, b+(C)=3b^+(C)=3 and b−(C)=12b^-(C)=12.

The proof occupies this section and the next three. Here we prove the following.

Proposition 2.2. In the situation of Theorem 2.1, one of the two pieces, WW, is negative definite and has oriented boundary YY. The groups H1(W;Z)H_{1}(W;\mathbb{Z}) and H1(C;Z)H_{1}(C;\mathbb{Z}) are zero, and the intersection form QWQ_{W} is isomorphic either to −E7-E_{7} or to ⟨−2⟩\langle-2\rangle.

Proof. Since H3(X;Z)=0H_{3}(X;\mathbb{Z})=0, the submanifold YY separates XX into two pieces. Let VV denote either of them. In the Mayer–Vietoris sequence, the groups H2(Y;Z)H_{2}(Y;\mathbb{Z}) and H1(X;Z)H_{1}(X;\mathbb{Z}) are zero, so H2(V;Z)H_{2}(V;\mathbb{Z}) injects into H2(X;Z)H_{2}(X;\mathbb{Z}) and is therefore free, while H1(V;Z)H_{1}(V;\mathbb{Z}) is a quotient of H1(Y;Z)H_{1}(Y;\mathbb{Z}) and is therefore finite. The exact sequence of the pair (V,Y)(V,Y) and Poincaré–Lefschetz duality then give

∣H1(Y;Z)∣=∣det⁡QV∣ ∣H1(V;Z)∣2.(2)|H_{1}(Y;\mathbb{Z})|=|\det Q_{V}|\,|H_{1}(V;\mathbb{Z})|^{2}. \tag*{(2)}

The left-hand side is 22, so H1(V;Z)=0H_{1}(V;\mathbb{Z})=0 and ∣det⁡QV∣=2|\det Q_{V}|=2. The form QVQ_{V} is even, because XX is spin.

The manifold YY is an LL-space and XX is symplectic, so one of the two pieces has b+=0b^{+}=0 [16 Theorem 1.2]. (If b+b^{+} were positive on both sides, then the Ozsváth–Szabó invariants of XX would factor through HFred(Y)=0HF_{\mathrm{red}}(Y)=0, and they would vanish, which they do not for a symplectic manifold.) We call this piece WW. It is negative definite, because QWQ_{W} is non-degenerate. To see which way the boundary of WW is oriented, we use the correction terms of YY. There are two spin⁡c\operatorname{spin}^{c} structures on YY, both of them spin, and their correction terms are

d(Y,s)=7/4and1/4.(3)d(Y,\mathfrak{s})=7/4 \quad\text{and}\quad1/4. \tag*{(3)}

See [1 Lemma 1.2], where the values are given for −Y-Y. If VV is a negative-definite spin manifold whose boundary is a rational homology sphere, and s\mathfrak{s} is the spin structure which the boundary inherits, then

b2(V)≤4d(∂V,s)b_{2}(V)\leq4d(\partial V,\mathfrak{s})

by [18 Theorem 9.6]. Since both correction terms of −Y-Y are negative, the oriented boundary of WW is YY and not −Y-Y, and the rank r=b2(W)r=b_{2}(W) is at most 77. It is at least 11, because ∣det⁡QW∣=2|\det Q_{W}|=2.

Let LL be the lattice −QW-Q_{W}, which is even and positive definite, of determinant 22. Its discriminant group has one non-zero element, on which the discriminant quadratic form takes a value qq which is either 1/21/2 or 3/23/2 modulo 22. The Milgram formula [14 Appendix 4] gives

e2πir/8=1+eπiq2,e^{2\pi i r/8}=\frac{1+e^{\pi i q}}{\sqrt{2}},

so r≡1(mod8)r\equiv1\pmod{8} if q=1/2q=1/2, and r≡7(mod8)r\equiv7\pmod{8} if q=3/2q=3/2. So either r=1r=1, in which case L=⟨2⟩L=\langle2\rangle, or r=7r=7 and q=3/2q=3/2. In the second case, the lattice L⊕A1L\oplus A_{1} has an even unimodular overlattice of index 22. This overlattice is positive definite of rank 88, and is therefore isomorphic to E8E_{8} [20 Chapter V]. In it, LL is the orthogonal complement of a root, so LL is isomorphic to E7E_{7}. □

To prove Theorem 2.1, it remains to show that QWQ_{W} cannot be ⟨−2⟩\langle-2\rangle.

The moduli spaces when NN has rank one

We suppose from now on that the second alternative in Proposition 2.2 occurs, and we write NN for WW. So XX is a K3 surface, decomposed as

X=N∪YC,QN≅⟨−2⟩,X=N\mathbin{\cup_{Y}}C,\qquad Q_{N}\cong\langle-2\rangle,

@ and our aim is to derive a contradiction. By Proposition 2.2, the first homology groups of NN and CC are zero. The group H2(N;Z)H_2(N;\mathbb{Z}) is generated by a class xx with x⋅x=−2x \cdot x=-2, and we have

χ(N)+σ(N)=1,b+(C)=3,b−(C)=18,χ(C)+σ(C)=7.(4)\chi(N)+\sigma(N)=1,\qquad b^{+}(C)=3,\qquad b^{-}(C)=18,\qquad\chi(C)+\sigma(C)=7. \tag*{(4)}

The class xx is primitive in H2(X;Z)H_2(X;\mathbb{Z}), because the intersection form of XX is even.

The bundle

Since xx is primitive and has square −2-2, there is an automorphism of the lattice H2(X;Z)H_2(X;\mathbb{Z}) which carries xx to e−fe-f, where ee and ff are the standard basis of a hyperbolic summand [8 Proposition 3.3(i)]. Let e′e' and f′f' be the standard basis of a second hyperbolic summand, orthogonal to the first, and let w∈H2(X;Z)w\in H^2(X;\mathbb{Z}) be the class dual to f+e′−3f′f+e'-3f'. Then

w[x]=1,w2=−6.(5)w[x]=1,\qquad w^2=-6. \tag*{(5)}

Let E→X\mathcal{E}\to X be the U(2)U(2) bundle with c1(E)=wc_1(\mathcal{E})=w and c2(E)=0c_2(\mathcal{E})=0, and let P=ad⁡(E)P=\operatorname{ad}(\mathcal{E}) be the associated SO(3)SO(3) bundle, so that

w2(P)=wmod⁡2,p1(P)[X]=−6.w_2(P)=w\mathbin{\operatorname{mod}}2,\qquad p_1(P)[X]=-6.

We adopt the usual conventions for such bundles, as in [11]: we consider connections AA in PP, and the gauge group G\mathcal{G} consists of the automorphisms of E\mathcal{E} of determinant one, so that an irreducible connection has stabilizer {±1}\{\pm1\} in G\mathcal{G}. We write κ\kappa for the action of a connection, normalized so that κ=−p1(P)/4\kappa=-p_1(P)/4 on a closed manifold. For the bundle PP we have κ=3/2\kappa=3/2, and the moduli space M(X,P)M(X,P) of anti-self-dual connections has formal dimension

8κ−3(1+b+(X))=0.8\kappa-3(1+b^{+}(X))=0.

The corresponding Donaldson invariant, which counts the points of a regular moduli space M(X,P)M(X,P) with signs, is

DXw(1)=±1.(6)D_X^w(1)=\pm1. \tag*{(6)}

This is the standard calculation for the K3 surface [11 Proposition 7.13]. We shall use only the fact that the number of points is odd.

Flat connections on YY

The restriction of ww to NN is the generator zz of H2(N;Z)≅ZH^2(N;\mathbb{Z})\cong\mathbb{Z} with z[x]=1z[x]=1, and its restriction to YY is the non-zero element of H2(Y;Z)=Z/2H^2(Y;\mathbb{Z})=\mathbb{Z}/2. The flat connections in P∣YP|_Y, modulo gauge transformations of determinant one, therefore correspond to the conjugacy classes of homomorphisms π1(Y)=O∗→U(2)\pi_1(Y)=O^*\to U(2) whose determinant is the non-trivial character ϵ:O∗→{±1}\epsilon:O^*\to\{\pm1\}.

Lemma 3.1. (a) There are two such conjugacy classes. They are represented by α=1⊕ϵ\alpha=1\oplus\epsilon and by β\beta, the pull-back of the irreducible two-dimensional representation of the symmetric group S3S_3 by the quotient maps O∗→S4→S3O^*\to S_4\to S_3.

(b) The rho-invariants of the corresponding flat connections in P∣YP|_Y are

ρY(α)=−4,ρY(β)=−173.\rho_Y(\alpha)=-4,\qquad\rho_Y(\beta)=-\frac{17}{3}.

(c) Their Chern–Simons invariants satisfy

cs⁡(β)−cs⁡(α)=−13(modZ).\operatorname{cs}(\beta)-\operatorname{cs}(\alpha)=-\frac{1}{3}\pmod{\mathbb{Z}}.

Here ρY(γ)\rho_Y(\gamma) is the Atiyah–Patodi–Singer rho-invariant of the flat bundle ad⁡γ\operatorname{ad}\gamma, and cs is the Chern–Simons invariant with values in R/Z\mathbb{R}/\mathbb{Z}. Our sign conventions for both are fixed by the way they enter the formulae (7) and (8) below. The proof of the lemma is a calculation with the character table of O∗O^*, and is given in § 4.4.

The flat connection α\alpha is reducible: because the characters 11 and ϵ\epsilon are distinct, its stabilizer in G/{±1}G/\{\pm1\} is a circle. The flat connection β\beta is irreducible. Both are non-degenerate, in that

H1(Y;ad⁡α)=H1(Y;ad⁡β)=0,H^1(Y;\operatorname{ad}\alpha)=H^1(Y;\operatorname{ad}\beta)=0,

because π1(Y)\pi_1(Y) is finite. We write hγh_\gamma for the dimension of the stabilizer of γ\gamma, so hα=1h_\alpha=1 and hβ=0h_\beta=0.

Remark 3.2. The holonomy group of the SO(3)SO(3) connection ad⁡α\operatorname{ad}\alpha has order two, so its centralizer in SO(3)SO(3) is a copy of O(2)O(2). The elements in the non-identity component of O(2)O(2) are automorphisms of P∣YP|_Y which do not lift to automorphisms of E∣YE|_Y of determinant one, and they play no role in what follows.

Moduli spaces on NN and CC

Let N+N^+ and C+C^+ be the manifolds obtained from NN and CC by attaching cylindrical ends [0,∞)×Y[0,\infty)\times Y and (−∞,0]×Y(-\infty,0]\times Y, with metrics which are products on the ends. For T>0T>0, let XTX_T be the Riemannian manifold

XT=N∪([−T,T]×Y)∪C,X_T=N\cup([-T,T]\times Y)\cup C,

which is a copy of XX containing a neck of length 2T2T.

Let VV be either NN or CC, and let γ\gamma be either α\alpha or β\beta. We write Mκ(V,γ)\mathcal{M}_\kappa(V,\gamma) for the moduli space of anti-self-dual connections in PP on V+V^+ which are asymptotic to γ\gamma on the end and have action κ\kappa. By the Atiyah–Patodi–Singer index theorem, the formal dimension of this moduli space is

d=8κ−32(χ(V)+σ(V))−12hγ+12ρ∂V(γ).(7)d=8\kappa-\frac{3}{2}\left(\chi(V)+\sigma(V)\right)-\frac{1}{2}h_\gamma+\frac{1}{2}\rho_{\partial V}(\gamma). \tag*{(7)}

where ρ∂N=ρY\rho_{\partial N}=\rho_Y and ρ∂C=−ρY\rho_{\partial C}=-\rho_Y.

On Mκ(V,γ)\mathcal{M}_\kappa(V,\gamma), the action κ\kappa is determined modulo Z\mathbb{Z} by γ\gamma. To find these residues, we consider reducible connections on NN. A reduction of P∣NP|_N to the circle has the form P=R⊕LP=\mathbb{R}\oplus L, where LL is a complex line bundle with c1(L)≡w(mod2)c_1(L)\equiv w\pmod2, so that c1(L)=kzc_1(L)=kz with kk odd. Because b+(N)=0b^+(N)=0 and H1(N;Z)=0H_1(N;\mathbb{Z})=0, each such line bundle carries an anti-self-dual connection on N+N^+ with curvature in L2L^2, unique up to gauge. Since z2=−1/2z^2=-1/2, the corresponding connection in PP has action

κ=−14c1(L)2=k2/8.\kappa=-\frac{1}{4}c_1(L)^2=k^2/8.

The reductions for kk and −k-k give the same connection in PP. So the reducible anti-self-dual connections on N+N^+ have action 1/81/8, 9/89/8, 25/825/8 and so on, and there is one of each. We write A0A_0 for the reducible connection with action 1/81/8. The limit of A0A_0 on the end is α\alpha, because α\alpha is the only reducible flat connection. So κ≡1/8\kappa\equiv1/8 on Mκ(N,α)\mathcal{M}_\kappa(N,\alpha), and from Lemma 3.1(c) we obtain

κ≡18  on Mκ(N,α),κ≡18−13=1924  on Mκ(N,β)(modZ).(8)\kappa\equiv\frac{1}{8}\ \text{ on }\mathcal{M}_\kappa(N,\alpha),\qquad \kappa\equiv\frac{1}{8}-\frac{1}{3}=\frac{19}{24}\ \text{ on }\mathcal{M}_\kappa(N,\beta)\pmod{\mathbb{Z}}. \tag*{(8)}

The residues on CC are obtained by subtracting these from 3/23/2. In the following table we list the pairs of non-negative values κN\kappa_N and κC\kappa_C allowed by (8) with κN+κC=3/2\kappa_N+\kappa_C=3/2, together with the formal dimensions dNd_N and dCd_C of MκN(N,γ)M_{\kappa_N}(N,\gamma) and MκC(C,γ)M_{\kappa_C}(C,\gamma) given by (7), (4) and Lemma 3.1(b).

γκNdNκCdCα1/8−311/82α9/853/8−6β19/24217/24−2(9)\begin{array}{c|cc|cc} \gamma& \kappa_N & d_N & \kappa_C & d_C \\ \alpha& 1/8 & -3 & 11/8 & 2 \\ \alpha& 9/8 & 5 & 3/8 & -6 \\ \beta& 19/24 & 2 & 17/24 & -2 \tag*{(9)} \end{array}

In each row, dN+dC+hγd_N+d_C+h_\gamma is zero, as it must be because M(X,P)M(X,P) has formal dimension zero.

The connection A0A_0 belongs to a moduli space of formal dimension −3-3. Its deformation complex decomposes according to the decomposition ad⁡P=R⊕L\operatorname{ad} P=\mathbb{R}\oplus L. The summand with coefficients in R\mathbb{R} has H0=RH^0=\mathbb{R}, the Lie algebra of the stabilizer of A0A_0, and its other cohomology groups are zero because b1(N)b_1(N) and b+(N)b^+(N) are zero. The summand with coefficients in LL is a complex of complex vector spaces, with H0=0H^0=0, and its index is −1-1. So HA02H^2_{A_0} has complex dimension at least one, with equality if and only if HA01=0H^1_{A_0}=0.

In the next proposition, the moduli spaces are defined using perturbations of the anti-self-duality equations on N+N^+ and C+C^+, which are of the type described in § 6.1 and are supported in compact subsets of the interiors of NN and CC. The same perturbations are used on XTX_T to define M(XT,P)M(X_T,P). Perturbing the equations does not affect the formal dimensions or the residues of κ\kappa. If the perturbations are small, then there is still exactly one reducible solution on N+N^+ with action 1/81/8, and we continue to call it A0A_0.

Proposition 3.3. The perturbations can be chosen so that the following hold.

(a) The moduli space M1/8(N,α)M_{1/8}(N,\alpha) consists of the single point [A0][A_0]. The cohomology group HA01H^1_{A_0} is zero, and HA02H^2_{A_0} is isomorphic to C\mathbb{C}, with the stabilizer of A0A_0 acting with weight one.

(b) The moduli spaces M3/8(C,α)M_{3/8}(C,\alpha) and M17/24(C,β)M_{17/24}(C,\beta) are empty. The moduli space M=M11/8(C,α)M=M_{11/8}(C,\alpha) is a compact smooth two-manifold. It consists of irreducible connections BB with HB2=0H^2_B=0.

(c) Let TiT_i be a sequence tending to infinity, and let [Ai][A_i] belong to M(XTi,P)M(X_{T_i},P). Then, after passing to a subsequence, there is a point [B][B] in MM such that the connections AiA_i converge to A0A_0 on compact subsets of N+N^+ and to BB on compact subsets of C+C^+. No action is lost in the limit: κ(A0)+κ(B)=3/2\kappa(A_0)+\kappa(B)=3/2.

Proof. In § 6.1 we show that the perturbations can be chosen so that the following hold. First, the connection A0A_0 has HA01=0H^1_{A_0}=0 and HA02≅CH^2_{A_0}\cong\mathbb{C}, and the irreducible parts of the moduli spaces on N+N^+ are regular (Lemma 5.1). Second, the moduli spaces on C+C^+ with κ≤3/2\kappa\leq3/2 contain no reducibles and are regular (Lemma 5.2). Third, the action of every solution on N+N^+ or C+C^+ is greater than −5/24-5/24; since the residues of κ\kappa are 1/81/8 and 19/2419/24 on NN, and 3/83/8 and 17/2417/24 on CC, the action is then non-negative.

A regular moduli space of negative formal dimension is empty. So the irreducible part of M1/8(N,α)M_{1/8}(N,\alpha) is empty, which proves (a), and the two moduli spaces on C+C^+ in the second and third rows of (9) are empty also. The moduli space MM is a smooth two-manifold. To see that it is compact, recall that a sequence in MM has a subsequence which converges weakly to a point of some Mκ′(C,γ)M_{\kappa'}(C,\gamma) with κ′≤11/8\kappa'\leq11/8. If κ′\kappa' is less than 11/811/8, then the difference is accounted for by bubbles, each of which carries action at least 11, and by anti-self-dual connections on the cylinder R×Y\mathbb{R} \times Y. The only possibilities allowed by the residues are M3/8(C,α)M_{3/8}(C,\alpha) and M17/24(C,β)M_{17/24}(C,\beta), and these are empty.

The proof of (c) is similar. A sequence [Ai][A_i] as in (c) has a subsequence which converges weakly to a pair of solutions on N+N^+ and C+C^+, belonging to moduli spaces MκN(N,γ)M_{\kappa_N}(N,\gamma) and MκC(C,γ′)M_{\kappa_C}(C,\gamma'), where κN\kappa_N and κC\kappa_C are non-negative and κN+κC≤3/2\kappa_N+\kappa_C\leq3/2. Equality holds only if there are no bubbles and no action is lost in the neck, in which case γ=γ′\gamma=\gamma'. Among the moduli spaces on C+C^+ with κC≤3/2\kappa_C\leq3/2, only MM can be non-empty. So κC=11/8\kappa_C=11/8 and κN≤1/8\kappa_N\leq1/8. Since κN\kappa_N is non-negative and congruent to 1/81/8 or 19/2419/24 modulo Z\mathbb{Z}, it is equal to 1/81/8, the limiting flat connection is α\alpha, and no action is lost. By (a), the limit on N+N^+ is A0A_0. □

Proof of Lemma 3.1

We refer to [9 Appendix A.4] for the character table of O∗O^*. The group has three irreducible two-dimensional representations: the defining representation QQ of O∗⊂SU(2)O^*\subset SU(2), the tensor product Q⊗ϵQ\otimes\epsilon, and β\beta. The first two have determinant 11, and β\beta has determinant ϵ\epsilon. A reducible representation with determinant ϵ\epsilon is the sum of two characters whose product is ϵ\epsilon, and so is 1⊕ϵ1\oplus\epsilon. This proves (a).

For (b), let ±θg\pm\theta_g be the arguments of the eigenvalues of g∈O∗g\in O^* in the representation QQ. With YY oriented as the boundary of the negative-definite plumbing, the rho-invariant is given by

ρY(γ)=148∑g≠1cot⁡2(θg2)(tr⁡ad⁡γ(g)−3).\rho_Y(\gamma)=\frac{1}{48}\sum_{g\neq1}\cot^2\left(\frac{\theta_g}{2}\right)(\operatorname{tr}\operatorname{ad}\gamma(g)-3).

The representations ad⁡α\operatorname{ad}\alpha and ad⁡β\operatorname{ad}\beta factor through the quotient S4S_4 of O∗O^*, and each element of S4S_4 is the image of two elements ±g\pm g of O∗O^*. In the following table, the third column is the sum of cot⁡2(θg/2)\cot^2(\theta_g/2) over the two elements of O∗O^* lying over an element of the given conjugacy class in S4S_4.

class in S4S_4size∑cot⁡2(θg/2)\sum\cot^2(\theta_g/2)tr⁡ad⁡α−3\operatorname{tr}\operatorname{ad}\alpha-3tr⁡ad⁡β−3\operatorname{tr}\operatorname{ad}\beta-3
(12)62−4-4−4-4
(123)810/310/30−3-3
(1234)66−4-4−4-4

Table 2.

The remaining conjugacy classes in S4S_4 make no contribution, because ad⁡α\operatorname{ad}\alpha and ad⁡β\operatorname{ad}\beta are trivial there. So

ρY(α)=−48−14448=−4,ρY(β)=−48−80−14448=−173.\rho_Y(\alpha)=\frac{-48-144}{48}=-4,\qquad\rho_Y(\beta)=\frac{-48-80-144}{48}=-\frac{17}{3}.

For (c), recall that H4(BO∗;Z)H^4(BO^*;\mathbb{Z}) is cyclic of order 48, generated by e=c2(Q)e=c_2(Q). We have c2(α)=0c_2(\alpha)=0. To find c2(β)c_2(\beta) we restrict to subgroups. On a Sylow 2-subgroup of O∗O^*, which is a generalized quaternion group of order 16, the representation β\beta is the sum of the trivial character and a non-trivial one, so c2(β)c_2(\beta) is zero modulo 16. On a cyclic subgroup of order 3, the representation β\beta coincides with QQ, so c2(β)c_2(\beta) is equal to ee modulo 3. Hence c2(β)=16ec_2(\beta)=16e. With the conventions of [9 Appendix B], the Chern–Simons invariant of QQ is −1/48-1/48. Since α\alpha and β\beta have the same determinant, it follows that

cs⁡(β)−cs⁡(α)=−1648=−13(modZ).\operatorname{cs}(\beta)-\operatorname{cs}(\alpha)=-\frac{16}{48}=-\frac{1}{3}\pmod{\mathbb{Z}}.

In (8), we have used the fact that if two connections on N+N^+ are asymptotic to γ\gamma and γ′\gamma', then the difference of their actions is congruent to cs⁡(γ)−cs⁡(γ′)\operatorname{cs}(\gamma)-\operatorname{cs}(\gamma') modulo Z\mathbb{Z}. For this, the Chern–Simons invariant is defined using the form ζ\zeta with dζ=(8π2)−1tr⁡(F∧F)\mathrm{d}\zeta=(8\pi^2)^{-1}\operatorname{tr}(F\wedge F), which is the normalization for which an anti-self-dual connection on a closed manifold has action κ=−p1/4\kappa=-p_1/4.

The parity argument

We continue with the notation of § 4. The gluing theorem of § 6 leads to the following description of the moduli space on XTX_T when TT is large. We write B∗(XT,P)\mathcal{B}^*(X_T,P) for the space of irreducible connections in PP modulo the gauge group G\mathcal{G}, and τ\tau for the perturbation of the equations on XTX_T, so that M(XT,P)\mathcal{M}(X_T,P) is the set of solutions of FA++τ(A)=0F_A^+ + \tau(A)=0 in B∗(XT,P)\mathcal{B}^*(X_T,P).

Proposition 4.1. If TT is sufficiently large then, after the addition of a further small perturbation to τ\tau, the following hold.

(a) There are an open subset UU of B∗(XT,P)\mathcal{B}^*(X_T,P), an oriented real two-plane bundle E→UE\to U, and a smooth bundle map ι\iota from EE to the bundle over UU whose fibre at [A][A] is Ω+(XT;ad⁡P)\Omega^+(X_T;\operatorname{ad}P), such that the space

Σ={([A],v):[A]∈U, v∈E[A], FA++τ(A)+ιA(v)=0}\Sigma=\{([A],v):[A]\in U,\ v\in E_{[A]},\ F_A^+ +\tau(A)+\iota_A(v)=0\}

is cut out transversely and is a compact surface, diffeomorphic to MM.

(b) The moduli space M(XT,P)\mathcal{M}(X_T,P) is regular and is contained in UU. So M(XT,P)\mathcal{M}(X_T,P) is the zero set of the section ss of E∣ΣE|_\Sigma given by s([A],v)=vs([A],v)=v, and ss vanishes transversely. The number of its zeros is odd.

The surface Σ\Sigma is a thickened moduli space, of the kind which arises whenever a moduli space is described by a finite-dimensional obstruction bundle. To say that it is cut out transversely means the following. Write σ(A)=FA++τ(A)\sigma(A)=F_A^+ +\tau(A). At a point ([A],v)([A],v) of Σ\Sigma, the linearization of the equation σ(A)+ιA(v)=0\sigma(A)+\iota_A(v)=0, together with the Coulomb gauge condition, is the operator

D(A,v):Ω1(XT;ad⁡P)⊕E[A]⟶Ω0(XT;ad⁡P)⊕Ω+(XT;ad⁡P),D_{(A,v)}:\Omega^1(X_T;\operatorname{ad}P)\oplus E_{[A]}\longrightarrow\Omega^0(X_T;\operatorname{ad}P)\oplus\Omega^+(X_T;\operatorname{ad}P),
(a,u)⟼(dA∗a,(dσ)A(a)+(∂aι)(v)+ιA(u)),(10)(a,u)\longmapsto(d_A^*a,(d\sigma)_A(a)+(\partial_a\iota)(v)+\iota_A(u)), \tag*{(10)}

and we require this operator to be surjective. Its kernel is then the tangent space to Σ\Sigma at ([A],v)([A],v).

The last statement in part (b) holds because the number of points in the moduli space, counted with signs, is the Donaldson invariant (6). On the other hand, the number of zeros of a transverse section of E∣ΣE|_\Sigma is given, modulo 2, by the second Stiefel–Whitney class. So Proposition 4.1 has the following consequence, in which [Σ][\Sigma] is the fundamental class with coefficients F2\mathbb{F}_2:

⟨w2(E),[Σ]⟩=1.(11)\langle w_2(E),[\Sigma]\rangle=1. \tag*{(11)}

Because G/{±1}\mathcal{G}/\{\pm1\} acts freely on the irreducible connections, there is a universal SO(3)SO(3) bundle over B∗(XT,P)×XT\mathcal{B}^*(X_T,P)\times X_T. Its restriction to Σ×XT\Sigma\times X_T carries a family of connections parametrized by Σ\Sigma. So the operators

DA=dA∗⊕dA+:Ω1(XT;ad⁡P)⟶Ω0(XT;ad⁡P)⊕Ω+(XT;ad⁡P),D_A=d_A^*\oplus d_A^+:\Omega^1(X_T;\operatorname{ad}P)\longrightarrow\Omega^0(X_T;\operatorname{ad}P)\oplus\Omega^+(X_T;\operatorname{ad}P),

for [A][A] in Σ\Sigma, form a family of elliptic operators over Σ\Sigma, which has an index ind⁡D\operatorname{ind}D in KO(Σ)KO(\Sigma).

Lemma 4.2. In KO(Σ)KO(\Sigma), we have

[TΣ]−[E∣Σ]=ind⁡D.(12)[T\Sigma]-[E|_\Sigma]=\operatorname{ind}D. \tag*{(12)}

Proof. The operators D(A,v)\mathcal{D}_{(A,v)} in (10) form a family of surjective Fredholm operators over Σ\Sigma, and the kernels form the bundle TΣT\Sigma. So the index of this family is [TΣ][T\Sigma]. On the other hand, D(A,v)\mathcal{D}_{(A,v)} differs from the operator (a,u)↦DAa(a,u) \mapsto D_Aa by terms which are compact: the derivative of the perturbation τ\tau is compact, and the terms involving ι\iota have finite rank. The index of the family (a,u)↦DAa(a,u) \mapsto D_Aa is ind⁡D+[E∣Σ]\operatorname{ind}D+[E|_\Sigma]. □

Proposition 4.3. Let XX be a K3 surface with any Riemannian metric, let BB be a compact space, and let P→B×XP \to B \times X be an SO(3)\mathrm{SO}(3) bundle carrying a family of connections on XX parametrized by BB. Then the index of the family of operators DA=dA∗⊕dA+D_A=d_A^*\oplus d_A^+ lies in the image of the map KSp(B)→KO(B)KSp(B)\to KO(B). In particular, its first and second Stiefel–Whitney classes are zero.

Proof. The index of the family does not depend on the Riemannian metric, so we may suppose that the metric is hyperkähler. The spin bundles S+S^+ and S−S^- of XX are quaternionic line bundles: we regard their fibres as right H\mathbb{H}-modules, and Clifford multiplication by one-forms and the Dirac operator are then H\mathbb{H}-linear. For the hyperkähler metric, the bundle S+S^+ is flat and trivial. Let ϕ\phi be a parallel section of S+S^+ of unit length. Clifford multiplication on ϕ\phi gives isomorphisms of real vector bundles

Λ1⟶S−,Λ0⊕Λ+⟶S+.\Lambda^1 \longrightarrow S^-, \qquad\Lambda^0\oplus\Lambda^+ \longrightarrow S^+.

and because ϕ\phi is parallel, these isomorphisms carry the operator d∗⊕d+d^*\oplus d^+ to the Dirac operator Γ(S−)→Γ(S+)\Gamma(S^-)\to\Gamma(S^+), up to constant factors. In the same way, the operator DAD_A becomes the Dirac operator coupled to the real vector bundle ad⁡P\operatorname{ad}P with the connection AA. This coupled Dirac operator is again H\mathbb{H}-linear. The quaternionic structure is the same for every member of the family, so the index of the family is the class of a virtual quaternionic vector bundle on BB. Finally, the real vector bundle which underlies a quaternionic vector bundle has w1=0w_1=0 and w2=0w_2=0. □

Proof of Theorem 2.1. By Proposition 2.2, we have only to exclude the case considered in § 4, in which QNQ_N is ⟨−2⟩\langle-2\rangle. In that case, let Σ\Sigma and EE be as in Proposition 4.1. By Proposition 4.3, the class ind⁡D\operatorname{ind}D in (12) has w1=0w_1=0 and w2=0w_2=0, and so the Stiefel–Whitney classes of TΣT\Sigma and E∣ΣE|_\Sigma are equal. Because EE is oriented, it follows that Σ\Sigma is orientable and that

⟨w2(E),[Σ]⟩=⟨w2(TΣ),[Σ]⟩.\langle w_2(E),[\Sigma]\rangle=\langle w_2(T\Sigma),[\Sigma]\rangle.

The right-hand side is the Euler number of Σ\Sigma modulo 2, which is zero because Σ\Sigma is a closed orientable surface. This contradicts (11). (If MM is empty, then Σ\Sigma is empty and (11) fails for that reason.) So QWQ_W is isomorphic to −E7-E_7, and the values of b+(C)b^+(C) and b−(C)b^-(C) follow. □

Regularity and gluing

In this section we prove the transversality statements which were used in the proof of Proposition 3.3, and we prove Proposition 4.1. The methods are not new. The technique for gluing in the presence of an obstruction, in which the equations are first solved modulo a finite-dimensional space associated with the small eigenvalues of the linearized operator, goes back to Taubes [22]; see also [3] and [5 Section 7.2]. For manifolds with cylindrical ends, we follow [4 Chapter 4]. Some constants in our argument grow polynomially with the length of the neck, while the error terms decay exponentially, and to deal with this one needs a slice theorem in which the constants are controlled by the lowest eigenvalue of the Laplacian dA∗dAd_A^*d_A. Results of that kind are also due to Taubes: see [23 Section 6] and [21 Lemma A.1], and the account in [6 Theorems 1.1 and 1.2]. The small eigenvalues of Dirac-type operators on manifolds with long necks, and the obstruction bundles to which they give rise in families, are discussed in [17]. The perturbations of the equations which we use are the holonomy perturbations of [10].

What is particular to the present situation is that the ends are rational homology spheres, on which the flat connections may be reducible. Moduli spaces of this sort are studied in detail in [13] and [2], and we refer to those papers at several points.

Perturbations and regularity

Let VV be an oriented Riemannian four-manifold with a cylindrical end [0,∞)×Y[0,\infty) \times Y, where YY is a rational homology sphere, and let P→VP \to V be an SO(3)SO(3) bundle, with the conventions of § 4. For a non-degenerate flat connection γ\gamma on the end, the moduli spaces Mκ(V,γ)M_\kappa(V,\gamma) are defined in the standard way [4 Chapter 4], using Sobolev spaces Lk2L_k^2 with k≥4k \ge4 and a small exponential weight on the end. For a solution AA, we write HA0H_A^0, HA1H_A^1 and HA2H_A^2 for the cohomology groups of its deformation complex, so that HA0H_A^0 is the Lie algebra of the stabilizer of AA and HA2H_A^2 is the cokernel of the linearized equations. The solution is regular if HA2=0H_A^2 = 0. The elements of HA2H_A^2 are represented by self-dual forms which decay exponentially on the end.

The equations are perturbed using holonomy perturbations, as in [10 Section 3]. In place of FA+=0F_A^+ = 0 we consider

σ(A)=FA++τ(A)=0,\sigma(A) = F_A^+ + \tau(A) = 0,

where τ\tau is a sum of terms of the following kind. One takes a ball BB in the interior of VV, a map q ⁣:S1×B→Vq\colon S^1 \times B \to V with q(1,x)=xq(1,x)=x, and a self-dual form ω\omega supported in BB. The corresponding term is obtained from the holonomy of AA around the loops q(⋅,x)q(\cdot,x), for xx in BB, by multiplying by ω\omega and projecting to Ω+(V;ad⁡P)\Omega^+(V;\operatorname{ad} P). The perturbations form a Banach space, they are supported in a compact subset of the interior of VV, and they are gauge-equivariant: if AA is compatible with a reduction P=R⊕LP = \mathbb{R} \oplus L, then τ(A)\tau(A) takes values in the summand R\mathbb{R}. Their use on manifolds with cylindrical ends, in the presence of reducible connections, is described in [13 Section 4] and [2 Section 3.2].

In [10], the map qq is required to be a submersion. We require more, namely that each map q(z,⋅)q(z,\cdot) is an embedding of BB. The derivative of τ\tau is then built from pull-backs by diffeomorphisms, so that it and its formal adjoint are bounded on Sobolev spaces of every order, negative orders included, which we need in the gluing argument. The usual transversality arguments apply without change to perturbations of this restricted kind.

A solution of the perturbed equations with action κ\kappa satisfies

κ=c0(∥FA−∥L22−∥FA+∥L22)≥−c0∥τ(A)∥L22,(13)\kappa= c_0\left(\lVert F_A^-\rVert_{L^2}^2 - \lVert F_A^+\rVert_{L^2}^2\right) \ge-c_0\lVert\tau(A)\rVert_{L^2}^2, \tag*{(13)}

where c0c_0 is a positive constant which depends only on the normalization of κ\kappa. The L2L^2 norm of τ(A)\tau(A) is bounded independently of AA, by a constant which is small when τ\tau is small. So for any ϵ>0\epsilon> 0, all solutions have κ>−ϵ\kappa> -\epsilon when τ\tau is sufficiently small.

If a solution AA is compatible with a reduction P=R⊕LP = \mathbb{R} \oplus L, then the linearization of the equations at AA in Coulomb gauge is the sum of a tangential operator, which acts on ordinary one-forms, and a normal operator DAND_A^N, which acts on one-forms with values in LL and is complex-linear. The circle in the stabilizer of AA acts on LL with weight one.

Lemma 5.1. Suppose that H1(V;Z)=0H_1(V;\mathbb{Z}) = 0 and b+(V)=0b^+(V) = 0. Let α\alpha be a flat connection on the end with H1(Y;ad⁡α)=0H^1(Y;\operatorname{ad}\alpha) = 0 whose stabilizer is a circle. Let P=R⊕LP = \mathbb{R} \oplus L be a reduction of PP which is compatible with α\alpha on the end, for which c1(L)c_1(L) is non-zero in H2(V;Q)H^2(V;\mathbb{Q}) and for which the normal operators have complex index −1-1. Then for all sufficiently small perturbations τ\tau,

there is exactly one solution A0A_0 compatible with this reduction, up to gauge. For a residual set of small perturbations, the normal operator DA0νD^\nu_{A_0} is injective, so that

HA01=0,HA02≅C,H^1_{A_0}=0,\qquad H^2_{A_0}\cong\mathbb{C},

and the irreducible solutions in the moduli spaces Mκ(V,γ)M_\kappa(V,\gamma) are regular, for all γ\gamma and for all κ\kappa below any given bound.

Proof. When τ=0\tau=0, the solutions compatible with the reduction are the anti-self-dual connections in LL with curvature in L2L^2, and there is exactly one of these up to gauge, because b+(V)=0b^+(V)=0 and H1(V;Z)=0H_1(V;\mathbb{Z})=0. The tangential operator at this solution has kernel H1(V;R)=0H^1(V;\mathbb{R})=0, and its cokernel consists only of the constants. So the solution persists when τ\tau is small, by the implicit function theorem. No others appear, because solutions in a line bundle cannot bubble and cannot lose action on the end; compare [13], Section 4.4.

The normal operators DτνD^\nu_\tau at these solutions are complex Fredholm operators of index −1-1. Among such operators, those whose kernel has complex dimension jj form a submanifold of real codimension 2j(j+1)2j(j+1), whose normal space at an operator with kernel KK and cokernel HH is Hom⁡C(K,H)\operatorname{Hom}_{\mathbb{C}}(K,H). Let KK and HH be the kernel and cokernel of D0νD^\nu_0. It is enough to show that the derivative of τ↦Dτν\tau\mapsto D^\nu_\tau at τ=0\tau=0 maps onto Hom⁡C(K,H)\operatorname{Hom}_{\mathbb{C}}(K,H). For surjectivity onto a finite-dimensional space is an open condition, so the map τ↦Dτν\tau\mapsto D^\nu_\tau is then transverse to all these submanifolds for τ\tau near zero, and by the Sard–Smale theorem DτνD^\nu_\tau is injective for a residual set of small τ\tau. The surjectivity is proved by the argument which is used in [13], Section 4.6 and [2], Section 3.5 to obtain transversality in the directions normal to the reducible connections. A non-zero element of KK is detected by the derivative of the holonomy around some loop in a compact subset of VV, because the flat line bundle L∣YL|_Y has no parallel sections. A non-zero element of HH cannot vanish on an open set. And the only constraint on the derivative of a holonomy perturbation in the directions normal to the reducible connections is equivariance under the circle, which for a real-linear map from KK to HH is the same as complex-linearity. The perturbations can be taken to vanish to first order at A0A_0 along the reducible locus, so that A0A_0 does not move.

The regularity of the irreducible solutions for generic τ\tau is proved as in [10], § 4 and § 6; see also [13], Section 4.6. □

Lemma 5.2. Suppose that H1(V;Z)=0H_1(V;\mathbb{Z})=0 and b+(V)>0b^+(V)>0, and that w2(P)w_2(P) is non-zero on the end. Then for a residual set of small perturbations τ\tau, the moduli spaces Mκ(V,γ)M_\kappa(V,\gamma) contain no reducible solutions and are regular, for all γ\gamma and for all κ\kappa below any given bound.

Proof. Since H1(V;F2)=0H^1(V;\mathbb{F}_2)=0, every reducible connection is compatible with a reduction P=R⊕LP=\mathbb{R}\oplus L. The class c1(L)c_1(L) reduces modulo 2 to w2(P)w_2(P), so it is non-zero on the end; and it is non-zero in H2(V;Q)H^2(V;\mathbb{Q}), because H2(V;Z)H^2(V;\mathbb{Z}) has no torsion. The solutions compatible with such a reduction are the zeros of a Fredholm section of index b1(V)−b+(V)b_1(V)-b^+(V), which is negative, so for generic τ\tau there are none; compare [13], Section 4.4. The regularity of the irreducible solutions is proved as before. □

These lemmas supply what was needed in the proof of Proposition 3.3. Lemma 5.1 applies to the manifold N+N^+ of § 4, with the reduction for which c1(L)=zc_1(L)=z: the formal dimension −3-3 of M1/8(N,α)M_{1/8}(N,\alpha) is the sum of −1-1, from the stabilizer, and twice the complex index of the normal operator. Lemma 5.2 applies to C+\mathbb{C}^+, where b1−b+=−3b_1-b^+=-3. The lower bound for the action comes from (13), with ϵ=5/24\epsilon=5/24.

The gluing theorem

Let XX be a closed, oriented four-manifold, decomposed as N∪YCN \cup_Y C along a rational homology sphere YY, and let P→XP \to X be an SO⁡(3)\operatorname{SO}(3) bundle, with the same conventions as before. We form N+N^+, C+C^+ and XTX_T as in § 4.3. We fix holonomy perturbations on N+N^+ and C+C^+, and we use their sum on XTX_T. Let α\alpha be a flat connection on P∣YP|_Y with H1(Y;ad⁡α)=0H^1(Y;\operatorname{ad}\alpha)=0 whose stabilizer Γα\Gamma_\alpha is a circle. We make the following hypotheses.

  1. A0A_0 is a reducible solution on N+N^+, asymptotic to α\alpha, with HA01=0H^1_{A_0}=0. The space H=HA02H=H^2_{A_0} is isomorphic to C\mathbb{C}, and the stabilizer ΓA0\Gamma_{A_0}, which is a circle, acts on it with weight one.

  1. MM is a compact two-manifold which is a union of components of the moduli space of solutions on C+C^+ asymptotic to α\alpha. It consists of irreducible, regular solutions.

  1. If TiT_i is a sequence tending to infinity, and [Ai][A_i] belongs to M(XTi,P)M(X_{T_i},P), then after passing to a subsequence there is a point [B][B] in MM such that the connections AiA_i converge to A0A_0 on compact subsets of N+N^+ and to BB on compact subsets of C+C^+, without loss of action.

Restriction to the end identifies ΓA0\Gamma_{A_0} with Γα\Gamma_\alpha. Let M~→M\widetilde{M} \to M be the moduli space of solutions framed at infinity, which is a principal Γα\Gamma_\alpha-bundle over MM, and let

E0=M~×ΓαHE_0=\widetilde{M}\times_{\Gamma_\alpha}H

be the associated complex line bundle.

Theorem 5.3. Suppose that hypotheses (i)–(iii) hold. Then there are constants T0T_0 and KK such that the following is true for all T≥T0T \ge T_0. Let ν\nu be a holonomy perturbation on XTX_T, supported in the complement of the neck, whose norm is at most e−KTe^{-KT}. Then assertion (a) of Proposition 4.1 holds for the equations on XTX_T defined by τ+ν\tau+\nu, with a diffeomorphism M→ΣM \to\Sigma which is covered by an isomorphism of oriented bundles E0→E∣ΣE_0 \to E|_\Sigma; and the moduli space M(XT,P)M(X_T,P) is contained in UU. For a residual set of such ν\nu, the moduli space M(XT,P)M(X_T,P) is regular.

Remark 5.4. The gluing parameter for the pair (A0,B)(A_0,B) is an element of Γα\Gamma_\alpha. Because ΓA0\Gamma_{A_0} maps isomorphically to Γα\Gamma_\alpha, the gluing parameter is absorbed by the action of ΓA0\Gamma_{A_0}, which is why Σ\Sigma is a copy of MM rather than of M~\widetilde{M}. In the situation of Remark 3.2, there is a second way to identify the two copies of ad⁡α\operatorname{ad}\alpha over the neck, using the non-identity component of O⁡(2)\operatorname{O}(2). The bundle which results is not PP: its second Stiefel–Whitney class differs from w2(P)w_2(P) by the image of the non-zero element of H1(Y;F2)H^1(Y;\mathbb{F}_2) under the coboundary map to H2(X;F2)H^2(X;\mathbb{F}_2), and this image is not zero because H1(N;F2)H^1(N;\mathbb{F}_2) and H1(C;F2)H^1(C;\mathbb{F}_2) are zero.

Proof of Theorem 5.3

We first take ν=0\nu=0.

Approximate solutions. Because α\alpha is non-degenerate, the connection A0A_0 approaches α\alpha exponentially on the end of N+N^+, and so do the solutions BB representing the points mm of MM on the end of C+C^+, uniformly in mm. We use cut-off functions to make A0A_0 and BB equal to α\alpha at distance greater than TT along the ends, and we join the results across the neck of XTX_T. This requires an identification of the two copies of α\alpha. The possible identifications form a copy of Γα\Gamma_\alpha, and because ΓA0\Gamma_{A_0} maps onto Γα\Gamma_\alpha, they all give gauge-equivalent connections. So we obtain connections AT(m)A_T(m), whose gauge-equivalence classes depend only on mm, with

∥σ(AT(m))∥Lk−12≤Ce−λT,(14)\|\sigma(A_T(m))\|_{L^2_{k-1}} \leq C e^{-\lambda T}, \tag*{(14)}

and with similar bounds for the derivatives with respect to mm. Here and below, CC, λ\lambda and qq denote positive constants which are independent of TT and mm. The manifolds XTX_T have uniformly bounded geometry, so the constants in the Sobolev and elliptic estimates do not depend on TT.

In the same way, the self-dual forms on N+N^+ which represent the elements of HH can be cut off and transferred to XTX_T. The result depends on the identification, through the action of ΓA0\Gamma_{A_0} on HH, and we obtain linear maps

ιm:(E0)m⟶Ω+(XT;ad⁡P).\iota_m:(E_0)_m\longrightarrow\Omega^+(X_T;\operatorname{ad}P).

The linearized equations. Let δm:Ω1(XT;ad⁡P)⟶Ω+(XT;ad⁡P)\delta_m:\Omega^1(X_T;\operatorname{ad}P)\longrightarrow\Omega^+(X_T;\operatorname{ad}P) be the derivative of σ\sigma at AT(m)A_T(m).

Lemma 5.5. There is a constant c>0c>0 such that, for all mm and all sufficiently large TT, the operator δmδm∗\delta_m\delta_m^* has exactly two eigenvalues in the interval [0,Ce−λT][0,Ce^{-\lambda T}], counted with multiplicity, and no other spectrum in [0,c][0,c]. If Πm\Pi_m is the projection onto the span of the two eigenvectors, then Πm∘ιm\Pi_m\circ\iota_m is an isomorphism from (E0)m(E_0)_m to the image of Πm\Pi_m.

Proof. The forms ιm(v)\iota_m(v) are annihilated by δm∗\delta_m^* up to an error of size Ce−λTCe^{-\lambda T}, which accounts for two small eigenvalues and for the last statement. If the lemma were false, there would be self-dual forms uiu_i on XTiX_{T_i}, with Ti→∞T_i\to\infty, which have L2L^2 norm 11, are orthogonal to the image of ιmi\iota_{m_i}, and have δmi∗ui→0\delta_{m_i}^*u_i\to0 in L2L^2. On the neck, where AT(m)A_T(m) is the pull-back of α\alpha, we identify self-dual forms with time-dependent one-forms on YY by η↦dt∧η+∗Yη\eta\mapsto dt\wedge\eta+*_{Y}\eta, and δδ∗\delta\delta^* becomes the operator

12(−∂t2+Δα),(15)\frac{1}{2}(-\partial_t^2+\Delta_\alpha), \tag*{(15)}

where Δα\Delta_\alpha is the Hodge Laplacian on Ω1(Y;ad⁡α)\Omega^1(Y;\operatorname{ad}\alpha). This Laplacian is bounded below by a positive constant, because H1(Y;ad⁡α)=0H^1(Y;\operatorname{ad}\alpha)=0, so the L2L^2 norm of uiu_i cannot escape into the neck. A subsequence therefore converges on compact subsets of N+N^+ and C+C^+ to L2L^2 solutions of the adjoint equations, not both zero. But the limit on C+C^+ is zero, because the points of MM are regular, and the limit on N+N^+ is an element of HH which is orthogonal to HH. □

The space H0(Y;ad⁡α)H^0(Y;\operatorname{ad}\alpha), which is not zero, does not enter in (15). It does affect the Laplacian dA∗dAd_A^*d_A on zero-forms, and so the Coulomb gauge condition. For A=AT(m)A=A_T(m), and for connections AA whose Lk2L_k^2 distance from AT(m)A_T(m) is at most T−qT^{-q}, we have

∥ξ∥L2≤CT∥dAξ∥L2,ξ∈Ω0(XT;ad⁡P).(16)\|\xi\|_{L^2}\leq CT\|d_A\xi\|_{L^2},\qquad\xi\in\Omega^0(X_T;\operatorname{ad}P). \tag*{(16)}

This is obtained from a Poincaré inequality on a compact subset of CC, where the connections representing the points of MM are irreducible, by integrating along the neck. So dA∗dAd_A^*d_A has an inverse GAG_A, whose norms as an operator between Sobolev spaces are bounded by powers of TT, and PA=1−dAGAdA∗P_A=1-d_AG_Ad_A^* is the projection onto the kernel of dA∗d_A^*. The powers of TT do no harm, because they multiply terms which are exponentially small. For example, if A=AT(m)A=A_T(m) and RmR_m denotes the operator ξ↦[σ(A),ξ]\xi\mapsto[\sigma(A),\xi], then the gauge-invariance of σ\sigma gives δmdA=Rm\delta_md_A=R_m, and so

δmPAδm∗=δmδm∗−RmGARm∗,(17)\delta_mP_A\delta_m^*=\delta_m\delta_m^*-R_mG_AR_m^*, \tag*{(17)}

where the last term has norm at most CTqe−2λTCTqe^{-2\lambda T}, by (14). With Lemma 5.5, this shows that (1−Πm)δm(1-\Pi_m)\delta_m has a right inverse SmS_m whose image lies in the kernel of dA∗d_A^* and whose norm, as an operator from Lk−12L_{k-1}^2 to Lk2L_k^2, is at most CTqCT^q.

The thickened moduli space. For mm in MM, consider the equation

σ(AT(m)+a)+ιm(v)=0(18)\sigma(A_T(m)+a)+\iota_m(v)=0 \tag*{(18)}

for a one-form aa in the image of SmS_m and an element vv of (E0)m(E_0)_m. Its linearization at (a,v)=(0,0)(a,v)=(0,0) is the map (a,v)↦δma+ιm(v)(a,v)\mapsto\delta_m a+\iota_m(v), which is invertible, with an inverse of norm at most CTqCT^q: this follows from the properties of SmS_m and from Lemma 5.5. The left-hand side of (18) is of size Ce−λTCe^{-\lambda T} at (0,0)(0,0), and the non-linear terms are quadratic, with constants bounded by powers of TT. So by the contraction mapping theorem, there is a QQ such that, for large TT, the equation has a unique solution (am,vm)(a_m,v_m) with

∥a∥Lk2+∣v∣≤T−Q.(19)\lVert a\rVert_{L^2_k}+|v|\leq T^{-Q}. \tag*{(19)}

The solution depends smoothly on mm.

The map m↦[AT(m)]m\mapsto[A_T(m)] embeds MM in B∗(XT,P)B^*(X_T,P), and up to an exponentially small error, the image of SmS_m is a complement in the Coulomb slice at AT(m)A_T(m) to the tangent space of the embedded surface. So the connections AT(m)+aA_T(m)+a, with aa in the image of SmS_m and ∥a∥Lk2<T−Q\lVert a\rVert_{L^2_k}<T^{-Q}, represent the points of an open neighborhood UU of the surface, each point once. This is an application of the slice theorem, in a form in which the size of the neighborhood is controlled by the lowest eigenvalue of dA∗dAd_A^*d_A, for which we have the bound (16); compare [23 Section 6] and [6 Theorem 1.2]. We define E→UE\to U and the map ι\iota of Proposition 4.1 by pulling back E0E_0 and the maps ιm\iota_m by the projection U→MU\to M. Then Σ\Sigma is the set of points ([AT(m)+am],vm)([A_T(m)+a_m],v_m). It is a copy of MM, it is cut out transversely, and the section ss is m↦vmm\mapsto v_m.

*The moduli space is contained in U.∗U.* We must show that, when TT is large, every element of M(XT,P)M(X_T,P) is represented by a connection AT(m)+aA_T(m)+a, with aa in the image of SmS_m and satisfying (19). For then (a,0)(a,0) is a solution of (18), and it coincides with (am,vm)(a_m,v_m) by uniqueness. Hypothesis (iii) does not give this directly, because the radius T−QT^{-Q} in (19) shrinks with TT.

Let [Ai][A_i] be elements of M(XTi,P)M(X_{T_i},P), with Ti→∞T_i\to\infty, which converge to A0A_0 and BB as in hypothesis (iii). Since no action is lost, AiA_i is close to α\alpha on the neck; and since α\alpha is non-degenerate, the difference a=Ai−αa=A_i-\alpha, in a suitable gauge, decays exponentially from both ends of the neck [4 Chapter 4]:

∥a(t)∥Lk2(Y)≤C(e−λ(t+Ti)+e−λ(Ti−t)),∣t∣≤Ti.(20)\lVert a(t)\rVert_{L^2_k(Y)}\leq C\left(e^{-\lambda(t+T_i)}+e^{-\lambda(T_i-t)}\right),\qquad|t|\leq T_i. \tag*{(20)}

Cutting AiA_i at t=0t=0 and extending the two halves by α\alpha, we obtain connections on N+N^+ and C+C^+ at which σ\sigma is of size Ce−λTiCe^{-\lambda T_i} in the weighted norms. The linearized operator at A0A_0 is injective with closed range, because HA01=0H^1_{A_0}=0, and the points of the compact space MM are regular. So these two connections lie within distance Ce−λTiCe^{-\lambda T_i} of A0A_0 and of a solution representing a point mim_i of MM. It follows that, after a gauge transformation, AiA_i is within distance CTiqe−λTiCT_i^q e^{-\lambda T_i} of ATi(mi)A_{T_i}(m_i), and this is much smaller than Ti−QT_i^{-Q} when ii is large.

The additional perturbation. If τ\tau is replaced by τ+ν\tau+\nu, where the norm of ν\nu is at most e−KTe^{-KT} and KK is large, then (14) still holds and the argument is unchanged. For a residual set of such ν\nu, the moduli space M(XT,P)M(X_T,P) is regular: perturbations supported in the complement of the neck are sufficient for this, because an element of the cokernel which vanishes there is zero, by unique continuation. This completes the proof of Theorem 5.3.

Proof of Proposition 4.1

In the situation of § 4, the stabilizer of α\alpha is a circle, and hypotheses (i), (ii) and (iii) are the three parts of Proposition 3.3. So Theorem 5.3 applies. It gives part (a) of Proposition 4.1 and the inclusion of M(XT,P)M(X_T,P) in UU. For a generic choice of the additional perturbation, the moduli space M(XT,P)M(X_T,P) is regular, and its points are then the transverse zeros of the section ss. For a generic choice it is also the case that there are no reducible solutions in PP, and that the moduli spaces for the bundles with smaller action, whose formal dimensions are −8-8 or less, are empty. The moduli space M(XT,P)M(X_T,P) is then compact, and the number of its points, counted with signs, is the Donaldson invariant (6), which is odd.

The cases D5D_5, D6D_6 and D7D_7

Let nn be 5, 6 or 7, and write YnY_n for YDnY_{D_n}. This manifold is the quotient of S3S^3 by the binary dihedral group

Γm=⟨a,b∣a2m=1, b2=am, bab−1=a−1⟩,m=n−2,\Gamma_m=\langle a,b\mid a^{2m}=1,\ b^2=a^m,\ bab^{-1}=a^{-1}\rangle,\qquad m=n-2,

which has order 4m4m. The group H1(Yn;Z)H_1(Y_n;\mathbb{Z}) is Z/4\mathbb{Z}/4 when nn is odd and Z/2⊕Z/2\mathbb{Z}/2\oplus\mathbb{Z}/2 when nn is even. In this section we prove the following.

Proposition 6.1. Let nn be 5, 6 or 7, and let NN be a negative-definite submanifold of a K3K3 surface XX, with oriented boundary YnY_n and with H1(N;Z)=0H_1(N;\mathbb{Z})=0. Then the intersection form of NN is not isomorphic to −Ln-L_n, where

L5=⟨4⟩,L6=A1⊕A1,L7=A3.L_5=\langle4\rangle,\qquad L_6=A_1\oplus A_1,\qquad L_7=A_3.

The proof follows § 4–§ 6 closely, and we shall concentrate on the points at which it differs. Suppose then that X=N∪YnCX=N\cup_{Y_n}C, where QNQ_N is isomorphic to −Ln-L_n and H1(N;Z)=0H_1(N;\mathbb{Z})=0.

The two pieces

The relation (2) holds for NN and for CC, with ∣H1(Yn;Z)∣=4\lvert H_1(Y_n;\mathbb{Z})\rvert=4 on the left. If H1(C;Z)H_1(C;\mathbb{Z}) were non-zero, it would have order 2 and QCQ_C would be unimodular. But QCQ_C is even, and its signature is n−20n-20, which is not divisible by 8. So H1(C;Z)=0H_1(C;\mathbb{Z})=0. As in § 4, we then have

χ(N)+σ(N)=1,b+(C)=3,χ(C)+σ(C)=7.(21)\chi(N)+\sigma(N)=1,\qquad b^+(C)=3,\qquad\chi(C)+\sigma(C)=7. \tag*{(21)}

Because the first homology groups of NN and CC are zero, the lattices H2(N;Z)H_2(N;\mathbb{Z}) and H2(C;Z)H_2(C;\mathbb{Z}) are primitive sublattices of H2(X;Z)H_2(X;\mathbb{Z}), each the orthogonal complement of the other, and the restriction map H2(X;Z)→H2(N;Z)H^2(X;\mathbb{Z})\to H^2(N;\mathbb{Z}) is surjective.

The bundle

We identify H2(N;Z)H^2(N;\mathbb{Z}) with the dual lattice Ln∗L_n^*, so that the square of a class cc in H2(N;Z)H^2(N;\mathbb{Z}) is −∣c∣2-\lvert c\rvert^2, where ∣c∣2\lvert c\rvert^2 denotes the norm in Ln∗L_n^*. We choose a class wNw_N in H2(N;Z)H^2(N;\mathbb{Z}) as follows. When n=5n=5, it is a generator of ⟨4⟩∗\langle4\rangle^*. When n=6n=6, it is z1z_1, where z1z_1 and z2z_2 are the basis of L6∗L_6^* dual to the basis of roots. When n=7n=7, we use the standard model of A3A_3 as the lattice of vectors in Z4\mathbb{Z}^4 with coordinate sum zero, and we take wN=w1w_N=w_1, where

wi=ei−14(e1+e2+e3+e4).w_i=e_i-\frac{1}{4}(e_1+e_2+e_3+e_4).

In each case ∣wN∣2=(n−4)/4\lvert w_N\rvert^2=(n-4)/4, and

the only elements of wN+2Ln∗ of norm (n−4)/4 are ±wN.(22)\text{the only elements of }w_N+2L_n^*\text{ of norm }(n-4)/4\text{ are }\pm w_N. \tag*{(22)}

This is clear when nn is 5 or 6. When n=7n=7, the vectors of norm 3/43/4 in A3∗A_3^* are the eight vectors ±wi\pm w_i, and for i≠ji\ne j the vectors (wi−wj)/2(w_i-w_j)/2 and (wi+wj)/2(w_i+w_j)/2 have norms 1/21/2 and 1/41/4, so they do not belong to A3∗A_3^*.

Lemma 6.2. There is a class ww in H2(X;Z)H^2(X;\mathbb{Z}) with w∣N=wNw|_N=w_N and w2=−6w^2=-6.

Proof. Let Λ=H2(C;Z)\Lambda=H_2(C;\mathbb{Z}). It is an even, indefinite lattice of rank 26−n26-n, and its discriminant form is the negative of that of H2(N;Z)H_2(N;\mathbb{Z}). When nn is odd, the discriminant group is cyclic of order 4, and the discriminant form takes the value 1/41/4 or 3/43/4 on a generator. When n=6n=6, it takes the values 1/21/2, 1/21/2 and 1 on the three non-zero elements. In neither case is there a non-zero isotropic element. By Meyer’s theorem, Λ\Lambda contains a primitive vector uu with u2=0u^2=0. Let dd be the positive generator of the ideal u⋅Λu\cdot\Lambda in Z\mathbb{Z}. Then u/du/d represents an isotropic element of the discriminant group, which is non-zero if d>1d>1. So d=1d=1, and there is a vector vv in Λ\Lambda with u⋅v=1u\cdot v=1. After adding a multiple of uu to vv, we may suppose that v2=0v^2=0, so that uu and vv span a hyperbolic plane in Λ\Lambda.

Now let w0w_0 be any class in H2(X;Z)H^2(X;\mathbb{Z}) which restricts to wNw_N. If we identify H2(X;Z)H^2(X;\mathbb{Z}) with H2(X;Z)H_2(X;\mathbb{Z}), then the classes which restrict to zero on NN are the elements of Λ\Lambda. So we may subtract a combination of uu and vv from w0w_0 to make it orthogonal to both, without changing its restriction to NN. The square of w0w_0 is even, and the class

w=w0+u−12(6+w02)vw=w_0+u-\frac{1}{2}(6+w_0^2)v

has the required properties. □

Let PP be the adjoint bundle of the U(2)U(2) bundle E\mathcal{E} with c1(E)=wc_1(\mathcal{E})=w and c2(E)=0c_2(\mathcal{E})=0. As in § 4, we have κ=3/2\kappa=3/2, the moduli space M(X,P)M(X,P) has formal dimension zero, and the Donaldson invariant is ±1\pm1.

Flat connections

The group H2(Yn;Z)H^2(Y_n;\mathbb{Z}) is the group of characters of Γm\Gamma_m, which has four elements. Let δ\delta be the character with

δ(a)=1,δ(b)=−1,\delta(a)=1,\qquad\delta(b)=-1,

and let χ\chi be the character which corresponds to the restriction of ww to YnY_n. When nn is odd, the class wNw_N generates Ln∗/Ln≅Z/4L_n^*/L_n\cong\mathbb{Z}/4, so χ\chi has order four and χ2=δ\chi^2=\delta. When n=6n=6, the character χ\chi is not trivial and is not δ\delta, because its self-linking number is non-zero, whereas δ\delta corresponds to the non-zero element of D6∗/D6D_6^*/D_6 which has self-linking number zero. In all three cases, therefore,

χ(a)=−1,\chi(a)=-1,

and the four characters of Γm\Gamma_m are 11, δ\delta, χ\chi and δχ\delta\chi.

Lemma 6.3. (a) Up to conjugacy, there are two homomorphisms Γm→U(2)\Gamma_m\to U(2) with determinant χ\chi. They are

α0=1⊕χ,α1=δ⊕δχ.\alpha_0=1\oplus\chi,\qquad\alpha_1=\delta\oplus\delta\chi.

(b) The rho-invariants of the corresponding flat connections are

ρYn(α0)=ρYn(α1)=−m.\rho_{Y_n}(\alpha_0)=\rho_{Y_n}(\alpha_1)=-m.

Proof. The irreducible two-dimensional representations of Γm\Gamma_m are the representations ρj\rho_j, for 1≤j≤m−11\leq j\leq m-1, given by

ρj(a)=(eπij/m00e−πij/m),ρj(b)=(0(−1)j10).\rho_j(a)=\begin{pmatrix} e^{\pi i j/m} & 0 \\ 0 & e^{-\pi i j/m} \end{pmatrix},\qquad \rho_j(b)=\begin{pmatrix} 0 & (-1)^j \\ 1 & 0 \end{pmatrix}.

See [9 Appendix A.2]. The determinant of ρj\rho_j is 1 or δ\delta, according as jj is odd or even, and is never χ\chi. A reducible representation with determinant χ\chi is the sum of two characters whose product is χ\chi, and there are two such pairs. This proves (a).

For (b), we use the formula for the rho-invariant from § 4.4, with 4m4m in place of 4848. For both α0\alpha_0 and α1\alpha_1, the character of the adjoint representation is 1+χ+χˉ1+\chi+\bar{\chi}. So the contribution of aja^j is zero when jj is even, and when jj is odd it is −4cot⁡2(πj/2m)-4\cot^2(\pi j/2m). Each of the 2m2m elements bajba^j has θ=π/2\theta=\pi/2, and the sum of tr⁡ad⁡αi−3\operatorname{tr}\operatorname{ad}\alpha_i-3 over these elements is −4m-4m. Since

∑0<j<2mj oddcot⁡2(πj2m)=m2−m,\sum_{\substack{0<j<2m\\ j\ \mathrm{odd}}}\cot^2\left(\frac{\pi j}{2m}\right)=m^2-m,

we obtain ρYn(αi)=(−4(m2−m)−4m)/4m=−m\rho_{Y_n}(\alpha_i)=(-4(m^2-m)-4m)/4m=-m.

The flat connections α0\alpha_0 and α1\alpha_1 have the same adjoint representation, but they are not equivalent under the gauge group G\mathcal{G} of determinant-one automorphisms, and we shall see that they are distinguished by the action. Both are reducible and non-degenerate. Their stabilizers in G/{±1}\mathcal{G}/\{\pm1\} are circles, because the characters 11 and χ\chi are distinct. (When n=6n=6, the holonomy group of ad⁡α0\operatorname{ad}\alpha_0 has order two, and Remarks 3.2 and 5.4 apply. When nn is odd, the holonomy group contains a rotation of order four, and its centralizer in SO(3)SO(3) is already a circle.)

The moduli spaces

Since α0\alpha_0 and α1\alpha_1 are both reducible, the residues of the action can be found from reducible connections on NN. A decomposition of E∣N\mathcal{E}|_N as L1⊕L2L_1\oplus L_2 is determined by the class c=c1(L1)−c1(L2)c=c_1(L_1)-c_1(L_2), which lies in wN+2H2(N;Z)w_N+2H^2(N;\mathbb{Z}). For each such cc there is a reducible anti-self-dual connection on N+N^+, unique up to gauge; the classes cc and −c-c give the same connection. Its action is ∣c∣2/4|c|^2/4, and its limit on the end is the sum of the two characters which are the restrictions to YnY_n of c1(L1)=(wN+c)/2c_1(L_1)=(w_N+c)/2 and c1(L2)=(wN−c)/2c_1(L_2)=(w_N-c)/2. Let

κ0=n−416.\kappa_0=\frac{n-4}{16}.

For c=wNc=w_N, the action is κ0\kappa_0 and the limit is 1⊕χ=α01\oplus\chi=\alpha_0. By (22), this is the only reducible connection with action κ0\kappa_0, and we call it A0A_0. Reducible connections with limit α1\alpha_1 are obtained by taking cc as in the following table, in which ω2=e1+e2−12(e1+e2+e3+e4)\omega_2=e_1+e_2-\frac{1}{2}(e_1+e_2+e_3+e_4).

nn567
cc3wN3w_Nz1+2z2z_1+2z_2ω1−2ω2\omega_1-2\omega_2
∣c∣2/4|c|^2/49/169/165/85/811/1611/16

Table 3.

In each case, the classes (wN±c)/2(w_N\pm c)/2 restrict to δ\delta and δχ\delta\chi on YnY_n, and the action is κ0+1/2\kappa_0+1/2. So

κ≡κ0on Mκ(N,α0),κ≡κ0+12on Mκ(N,α1)(modZ).(23)\kappa\equiv\kappa_0\quad\text{on }M_\kappa(N,\alpha_0),\qquad \kappa\equiv\kappa_0+\frac{1}{2}\quad\text{on }M_\kappa(N,\alpha_1)\pmod{\mathbb{Z}}. \tag*{(23)}

From (7), (21) and Lemma 6.3, we obtain the following table, which takes the place of (9).

γ\gammaκN\kappa_NdNd_NκC\kappa_CdCd_C
α0\alpha_0κ0\kappa_0−3-33/2−κ03/2-\kappa_022
α1\alpha_1κ0+1/2\kappa_0+1/2111−κ01-\kappa_0−2-2
α0\alpha_0κ0+1\kappa_0+1551/2−κ01/2-\kappa_0−6-6

Table 24.

Lemma 5.1 applies to N+N^+, with the reduction for which c=wNc=w_N, and Lemma 5.2 applies to C+C^+. We choose the perturbations so small that all solutions have κ>−κ0\kappa>-\kappa_0, which by (23) implies that κ≥0\kappa\geq0. The three parts of Proposition 3.3 then hold, with the following changes. In (a), the moduli space is Mκ0(N,α0)\mathcal{M}_{\kappa_0}(N,\alpha_0). In (b), the moduli spaces on C+C^+ in the second and third rows of (24) are empty, and the compact two-manifold is

M=M3/2−κ0(C,α0).M = \mathcal{M}_{3/2-\kappa_0}(C,\alpha_0).

The proofs are the same. For the compactness of MM and for part (c), we note that an anti-self-dual connection on R×Yn\mathbb{R} \times Y_n which runs between α0\alpha_0 and α1\alpha_1 has action at least 1/21/2, while a bubble, or a non-trivial anti-self-dual connection on the cylinder with the same limit at both ends, has action at least 1. So the only moduli spaces on C+C^+ which could arise from weak limits are the ones in the table.

The hypotheses of Theorem 5.3 are therefore satisfied, and Proposition 4.1 holds in the present situation. The argument of § 5 then gives a contradiction, as before, and Proposition 6.1 is proved.

The remaining cases

In this section we complete the proof of Theorem 1.1. Throughout, RR is a connected subdiagram of the Dynkin diagram E8E_8, and NN is a submanifold of a K3 surface XX which satisfies the hypotheses of the theorem: it is negative definite, its oriented boundary is YRY_R, and H1(N;Z)=0H_1(N;\mathbb{Z})=0. We write CC for the closure of the complement of NN in XX.

Proposition 7.1. The group H2(N;Z)H_2(N;\mathbb{Z}) is free, and H3(N;Z)H_3(N;\mathbb{Z}) is zero. The intersection form QNQ_N either is isomorphic to −R-R, or is one of the four forms listed in (1). The group H1(C;Z)H_1(C;\mathbb{Z}) is zero, except possibly when R=A7R=A_7, in which case it is either zero or Z/2\mathbb{Z}/2.

Proof. The complement CC is connected, and both NN and CC are spin. As in the proof of Proposition 2.2, for V=NV=N or CC the group H2(V;Z)H_2(V;\mathbb{Z}) is free, the group H1(V;Z)H_1(V;\mathbb{Z}) is finite, and

∣H1(YR;Z)∣=∣det⁡QV∣∣H1(V;Z)∣2.(24)\left|H_1(Y_R;\mathbb{Z})\right|=\left|\det Q_V\right|\left|H_1(V;\mathbb{Z})\right|^2. \tag*{(24)}

The boundary map H4(X;Z)→H3(YR;Z)H_4(X;\mathbb{Z})\to H_3(Y_R;\mathbb{Z}) in the Mayer–Vietoris sequence is an isomorphism, so H3(N;Z)H_3(N;\mathbb{Z}) and H3(C;Z)H_3(C;\mathbb{Z}) are zero. From (24) we see that L=−QNL=-Q_N is an even, positive-definite lattice whose determinant is ∣H1(YR;Z)∣=det⁡R\left|H_1(Y_R;\mathbb{Z})\right|=\det R. Let rr be its rank.

Correction terms. The spin⁡c\operatorname{spin}^{c} structures on YRY_R are indexed by the elements λ\lambda of R∗/RR^*/R, in such a way that the spin structures correspond to the elements with 2λ=02\lambda=0, and the formula of [19 Corollary 1.5] for the boundary of a negative-definite plumbing gives

d(YR,sλ)=rk⁡R4−min⁡v∈R∣λ+v∣2.(25)d(Y_R,\mathfrak{s}_\lambda)=\frac{\operatorname{rk}R}{4}-\min_{v\in R}\left|\lambda+v\right|^2. \tag*{(25)}

From the minimal norms of the cosets of the root lattices in their duals, we obtain the correction terms of the spin structures:

RRd(YR,s)d(Y_R,\mathfrak{s}), s\mathfrak{s} spin
AnA_n, nn evenn/4n/4
AnA_n, nn oddn/4n/4, −1/4-1/4
DnD_n, nn oddn/4n/4, n/4−1n/4-1
DnD_n, nn evenn/4n/4, n/4−1n/4-1, 00, 00
E6E_63/23/2
E7E_77/47/4, 1/41/4
E8E_822

Table 27.

(For DnD_n the four cosets have minimal norms 00, 11, n/4n/4 and n/4n/4, and the last two have order two only when nn is even. For AnA_n the coset corresponding to j∈Z/(n+1)j \in\mathbb{Z}/(n+1) has minimal norm j(n+1−j)/(n+1)j(n+1-j)/(n+1), and it has order two when j=(n+1)/2j=(n+1)/2.)

Let ss be the spin structure which YRY_R inherits from NN. Then

r≤4d(YR,s),r≡4d(YR,s)(mod8).(26)r \le4d(Y_R,s), \qquad r \equiv4d(Y_R,s) \pmod{8}. \tag*{(26)}

The inequality is [18 Theorem 9.6], and the congruence follows from the grading shift of the map which the punctured manifold NN induces on HF∞HF^\infty. If r=0r=0, then NN is a rational homology ball and d(YR,s)=0d(Y_R,s)=0.

The diagrams E6E_6, E7E_7 and E8E_8. From (27) and (26), the rank rr is 66 for E6E_6, it is 11 or 77 for E7E_7, and it is 00 or 88 for E8E_8. In the last case, r=0r=0 is excluded because the correction term is not zero, and LL is an even unimodular lattice of rank 88, so it is isomorphic to E8E_8. For E6E_6, the discriminant form of LL is the negative of the discriminant form of A2A_2. So L⊕A2L \oplus A_2 has an even unimodular overlattice, which is isomorphic to E8E_8, and LL is the orthogonal complement of a sublattice A2A_2 in E8E_8, which is E6E_6. The case of E7E_7 was treated in Proposition 2.2.

The diagrams DnD_n. The determinant of LL is 44, by (24) and the hypothesis that H1(N;Z)H_1(N;\mathbb{Z}) is zero, so rr is not zero. (For D4D_4 and D6D_6, the correction terms alone do not exclude r=0r=0.) From (27) and (26), we then have r=4r=4 for D4D_4, and

r∈{1,5},r∈{2,6},r∈{3,7}r \in\{1,5\}, \qquad r \in\{2,6\}, \qquad r \in\{3,7\}

for D5D_5, D6D_6 and D7D_7 respectively. An even positive-definite lattice of determinant 44 is ⟨4⟩\langle4\rangle if its rank is 11, and is A1⊕A1A_1 \oplus A_1 if its rank is 22. When r≥3r \ge3, the discriminant form of LL is determined by the linking form of YRY_R and the value of rr modulo 88, through Milgram’s formula, and it is the negative of the discriminant form of the lattice L′L' in the following table.

rr3344556677
L′L'D5D_5D4D_4A3A_3A1⊕A1A_1 \oplus A_1⟨4⟩\langle4\rangle
LLA3A_3D4D_4D5D_5D6D_6D7D_7

Table 1.

So L⊕L′L \oplus L' has an even unimodular overlattice, isomorphic to E8E_8, in which LL is the orthogonal complement of L′L'. The Weyl group of E8E_8 acts transitively on the sublattices of E8E_8 of each of the types A3A_3, D4D_4 and D5D_5, on the pairs of orthogonal roots, and on the vectors of norm 44. So the complement can be computed in the standard model, in which E8E_8 consists of the vectors xx in Z8∪(Z+12)8\mathbb{Z}^8 \cup(\mathbb{Z}+\frac{1}{2})^8 with ∑xi\sum x_i even. A sublattice DkD_k in the first kk coordinates, where D3=A3D_3=A_3, has complement D8−kD_{8-k} in the remaining coordinates. The orthogonal roots e1−e2e_1-e_2 and e1+e2e_1+e_2 have complement D6D_6, and the vector 2e12e_1 has complement D7D_7. This gives the last row of the table.

The diagrams AnA_n. When nn is odd, the spin structure with correction term −1/4-1/4 cannot be the one inherited from NN. So d(YR,s)=n/4d(Y_R,s)=n/4, and from (26) and the fact that n≤7n \le7 we obtain r=nr=n. To identify the lattice, let VV be the disk bundle over the two-sphere with Euler number −(n+1)-(n+1). Its boundary is −YAn-Y_{A_n}, so we can form the closed manifold Z=N∪VZ=N\cup V. It is negative definite and has H1(Z;Z)=0H_1(Z;\mathbb{Z})=0, so by Donaldson’s theorem its intersection form is diagonalizable. Changing the sign, we identify H2(Z;Z)H_2(Z;\mathbb{Z}) with the standard lattice Zn+1\mathbb{Z}^{n+1}. Let v∈Zn+1v\in\mathbb{Z}^{n+1} be the class of the zero section of VV, so that ∣v∣2=n+1|v|^2=n+1. The vector vv is primitive, because the quotient of H2(Z;Z)H_2(Z;\mathbb{Z}) by H2(V;Z)H_2(V;\mathbb{Z}) is a subgroup of H2(N,YAn;Z)H_2(N,Y_{A_n};\mathbb{Z}), which is free. The lattice LL is contained in the orthogonal complement of vv, and it is equal to it because the two have the same rank and determinant.

Because vv is primitive, the reduction of v⊥v^{\perp} modulo 2 is the kernel of the linear form u↦v⋅uu \mapsto v \cdot u on (Z/2)n+1(\mathbb{Z}/2)^{n+1}. Because v⊥v^{\perp} is even, the linear form u↦∣u∣2u \mapsto|u|^2, which is not zero, vanishes on this kernel. So the two linear forms are equal, which means that every coordinate of vv is odd. Since ∣v∣2=n+1|v|^2=n+1, every coordinate is ±1\pm1, and v⊥v^{\perp} is isomorphic to AnA_n.

The complement. By (25), the group H1(C;Z)H_1(C;\mathbb{Z}) is zero when ∣H1(YR;Z)∣|H_1(Y_R;\mathbb{Z})| is square-free. When ∣H1(YR;Z)∣=4|H_1(Y_R;\mathbb{Z})|=4, which is the case for A3A_3 and for DnD_n, the group H1(C;Z)H_1(C;\mathbb{Z}) could only have order 2, and QCQ_C would then be even and unimodular. But the signature of QCQ_C is r−16r-16, with 1≤r≤71\leq r\leq7, and this is not divisible by 8. In the remaining case, R=A7R=A_7, the relation (25) allows H1(C;Z)H_1(C;\mathbb{Z}) to have order 1 or 2.

Proof of Theorem 1.1. By Proposition 7.1, either QNQ_N is isomorphic to −R-R or we are in one of the four cases of [1]. The first of these is excluded by Theorem 2.1, in which NN is the piece called WW, and the other three are excluded by Proposition 6.1.

For R=D5R=D_5 and R=D7R=D_7, as for R=E7R=E_7, the hypotheses of Theorem 1.1 can be removed.

Theorem 7.2. Let nn be 5 or 7, and suppose that Yn=YDnY_n=Y_{D_n} is smoothly embedded in a K3K3 surface XX. Then YnY_n separates XX into two pieces, X=W∪YnCX=W\cup_{Y_n}C, where WW has oriented boundary YnY_n and is negative definite. The groups H1(W;Z)H_1(W;\mathbb{Z}) and H1(C;Z)H_1(C;\mathbb{Z}) are zero, and the intersection form of WW is isomorphic to −Dn-D_n.

Proof. As in the proof of Proposition 2.2, the manifold YnY_n is an LL-space, and one of the two pieces, WW, is negative definite. By (27), the correction terms of the two spin structures on YnY_n are n/4n/4 and n/4−1n/4-1. Both are positive, so the inequality of [18], Theorem 9.6 shows that the oriented boundary of WW is YnY_n and that b2(W)≤nb_2(W)\leq n. By (25), the group H1(W;Z)H_1(W;\mathbb{Z}) has order 1 or 2. If the order were 2, then QWQ_W would be even and unimodular, and its rank would be divisible by 8. The rank is at most 7, so WW would be a rational homology ball. The correction term of the spin structure inherited from WW would then be zero, which it is not. So H1(W;Z)=0H_1(W;\mathbb{Z})=0, and Theorem 1.1 shows that QWQ_W is isomorphic to −Dn-D_n. The group H1(C;Z)H_1(C;\mathbb{Z}) is zero by Proposition 7.1.

References

  1. [1]A. Daemi, T. Lidman and M. Miller Eismeier, 3-manifolds without any embedding in symplectic 4-manifolds, Geom. Topol. 28 (2024), 3357–3372.DOI
  2. [2]A. Daemi and M. Miller Eismeier, Instantons and rational homology spheres, preprint, arXiv:2210.14071. (Section numbers refer to version 2.)arxiv.org/abs/2210.14071
  3. [3]S. K. Donaldson, Connections, cohomology and the intersection forms of 4-manifolds, J. Differential Geom. 24 (1986), 275–341.DOI
  4. [4]S. K. Donaldson, Floer homology groups in Yang–Mills theory, Cambridge Tracts in Mathematics, vol. 147, Cambridge University Press, 2002.DOI
  5. [5]S. K. Donaldson and P. B. Kronheimer, The geometry of four-manifolds, Oxford Mathematical Monographs, Oxford University Press, 1990.
  6. [6]P. M. N. Feehan, Critical-exponent Sobolev norms and the slice theorem for the quotient space of connections, Pacific J. Math. 200 (2001), 71–118.
  7. [7]P. M. N. Feehan and T. G. Leness, The SO(3) monopole cobordism and superconformal simple type, Advances in Mathematics (2019), article 106817; arXiv:1408.5307, version 5. arxiv.org/abs/1408.5307. See Equation (1.2), Theorem 1.1, Hypothesis 2.5, and Remark 2.8.
  8. [8]V. Gritsenko, K. Hulek and G. K. Sankaran, Abelianisation of orthogonal groups and the fundamental group of modular varieties, J. Algebra 322 (2009), 463–478.arxiv.org/abs/0810.1614
  9. [9]G. O. Helle, Equivariant instanton Floer homology and calculations for the binary polyhedral spaces, preprint, arXiv:2203.09471.arxiv.org/abs/2203.09471
  10. [10]P. B. Kronheimer, Four-manifold invariants from higher-rank bundles, J. Differential Geom. 70 (2005), 59–112.DOI
  11. [11]P. B. Kronheimer and T. S. Mrowka, Embedded surfaces and the structure of Donaldson’s polynomial invariants, J. Differential Geom. 41 (1995), 573–734.DOI
  12. [12]M. Mariño, G. Moore, and G. Peradze, Four-manifold geography and superconformal symmetry, Mathematical Research Letters 6 (1999), 429–437; arXiv:math/9812042. arXiv.org/abs/math/9812042.DOI
  13. [13]M. Miller Eismeier, Equivariant instanton homology, preprint, arXiv:1907.01091. (Section numbers refer to version 3.)
  14. [14]J. Milnor and D. Husemoller, Symmetric bilinear forms, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 73, Springer, 1973.DOI
  15. [15]L. Moser, Elementary surgery along a torus knot, Pacific J. Math. 38 (1971), 737–745.DOI
  16. [16]A. Mukherjee, A note on embeddings of 3-manifolds in symplectic 4-manifolds, preprint, arXiv:2010.03681.
  17. [17]L. I. Nicolaescu, On the Cappell–Lee–Miller gluing theorem, Pacific J. Math. 206 (2002), 159–185.DOI
  18. [18]P. Ozsváth and Z. Szabó, Absolutely graded Floer homologies and intersection forms for four-manifolds with boundary, Adv. Math. 173 (2003), 179–261.DOI
  19. [19]P. Ozsváth and Z. Szabó, On the Floer homology of plumbed three-manifolds, Geom. Topol. 7 (2003), 185–224.DOI
  20. [20]J.-P. Serre, A course in arithmetic, Graduate Texts in Mathematics, vol. 7, Springer, 1973.DOI
  21. [21]C. H. Taubes, Path-connected Yang–Mills moduli spaces, J. Differential Geom. 19 (1984), 337–392.DOI
  22. [22]C. H. Taubes, Self-dual connections on 4-manifolds with indefinite intersection matrix, J. Differential Geom. 19 (1984), 517–560.DOI
  23. [23]C. H. Taubes, A framework for Morse theory for the Yang–Mills functional, Invent. Math. 94 (1988), 327–402.DOI

Paper details

vlnc:W_uu67gnm3
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

OpenAI ChatGPT 6, Anthropic Claude Fable 5.2

View statement
Looking for related papers…
Contents