Harmonic sections in sphere bundles, normal neighborhoods of reduction loci, and instanton moduli spaces on definite 4-manifolds (Q2464832)

When the Seiberg-Witten invariants were discovered in 1994, a standard question was, `is Donaldson theory dead?' The point was that it seemed possible to reprove the results of Donaldson theory using the Seiberg-Witten equations and the proofs seemed easier. This is because the moduli space considered in the Seiberg-Witten case is compact. Now almost all work in low-dimensional topology has shifted away from Donaldson theory and even away from Seiberg-Witten theory. In this paper, Andrei Teleman shows that Donaldson theory is still alive and well. In Donaldson theory one associates the moduli space of anti-self dual connections to a given \(4\)-manifold. When things work right one can show that this is a signed collection of points and the number of such is an invariant of the underlying \(4\)-manifold. Things rarely go right. One can compute the expected dimension of the moduli space. When this number is positive one can add additional constraints (intersect divisors in the moduli space coming from homology classes in the manifold) to arrive at a zero dimensional space and then count the points. The hope is that these moduli are at least finite-dimensional compact manifolds so that the intersection numbers of the divisors will remain unchanged under deformations of the initial geometric data. Even though the moduli space is non-compact in general the way in which it fails to be compact is well-understood and it is usually possible to define good compactifications of it. There is a second problem coming from the fact that the moduli space is the quotient of a space by a group. This means that singularities (reducible connections) appear whenever there are points with non-trivial stabilizer subgroups. Donaldson avoided this problem by restricting to manifolds with \(b_+^2>1\). Since the codimension of the space of reducibles is \(b_+^2\) this condition means that all reducibles may be avoided by a generic choice of metric and two generic metrics can be connected by a \(1\)-parameter family of metrics avoiding the reducibles, so intersection theory will work as expected. The theory was later extended to cover the \(b_+^2\) case where there are `walls' separating `chambers' of good metrics and one can study how the invariant changes when the walls are crossed and look for canonical chambers. In this paper, Teleman constructs invariants for negative-definite (\(b_+^2=0\)) \(4\)-manifolds by a careful study of the ASD moduli space. The first ingredient is a theorem proving that there is a connected, dense open set of metrics for which all reductions in the Uhlenbeck compactification are simultaneously regular. The proof takes a few pages of analysis. (The old regularity result only showed that there was a dense set of good metrics.) The next section of the paper gives a careful description of a neighborhood of the reducible connections in the moduli space. The novel ingredient here is that the reducibles are not isolated. In a neighborhood of the reduction locus the space of instantons can be analyzed via a system of equations for an abelian connection that is very similar in form to the Seiberg-Witten equations. Using the new regularity result and new model of the reducibles, Teleman defines invariants when the rank \(2\)-bundles with certain Chern classes are topologically non-decomposable. He also defines invariants when the bundle \(E\) splits topologically and satisfies \(4c_2(E)-c_1(E)^2\leq 3\) and \(c_1(E)\in 2H^2(X,\mathbb{Z})+\text{Tor}\).