Donaldson's theorem, Heegaard Floer homology, and knots with unknotting number one (Q2445934)

A knot \(K\) in the 3-sphere is said to have unknotting number one if it may be converted to the unknot by a single crossing change. Here a crossing change is a local modification in a small ball whose intersection with the knot consists of two arcs; it may also be realised by changing the overcrossing and undercrossing strands in a double point of some diagram of the knot. It follows by what is often referred to as Montesinos' trick that the double cover \(\Sigma(K)\) of \(S^3\) branched along the knot is given by half-integer surgery on some knot \(C\) in \(S^3\); to see this, note that the double cover of a 3-ball branched along two arcs is a solid torus. Thus a common way to obstruct unknotting number one is to apply some obstruction to \(\Sigma(K)\) being given as half-integer Dehn surgery on a knot. There are two such obstructions coming from gauge theory (broadly construed) which we describe below; the main result of this paper is a beautiful obstruction which combines these two. If an oriented 3-manifold \(Y\) is given as \((2n-1)/2\)-surgery on a knot \(C\), there is a standard procedure for converting this to an integer-framed surgery on a link. In this case the link consists of the knot \(C\) together with a meridian of \(C\), and the framings on these two components are \(n\) and \(2\) respectively. The trace of the surgery is a 4-manifold \(W\) bounded by \(Y\) with a positive-definite intersection form represented by the matrix \(\begin{pmatrix} n&-1\\-1&2\end{pmatrix}\). If \(-Y\) is also known to bound a smooth positive-definite 4-manifold \(X\), then gluing \(X\) and \(W\) along \(Y\) results in a closed positive-definite manifold whose intersection form is diagonalisable by \textit{S. Donaldson}'s celebrated theorem [J. Differ. Geom. 26, 397--428 (1987; Zbl 0683.57005)]. This gives a lattice-theoretic obstruction as follows: if \(Y=-\partial X\) is \((2n-1)/2\)-surgery on a knot, then the intersection lattice of \(X\) embeds as a codimension 2 sublattice of the integer lattice \(\mathbb{Z}^{m+2}\), with vectors \(v\) and \(w\) in its orthogonal complement which satisfy \(v.v=2\), \(v.w=-1\), and \(w.w=n\). A further obstruction comes from Heegaard Floer homology. The Heegaard Floer invariants of Dehn surgery on a knot \(C\) in \(S^3\) may be expressed in terms of the knot Floer invariants of \(C\) via the mapping cone formula of \textit{P. S. Ozsváth} and \textit{Z. Szabó} [Algebr. Geom. Topol. 11, No. 1, 1--68 (2011; Zbl 1226.57044)]. This imposes restrictions on those invariants, and in particular the Heegaard Floer correction terms of \(p/q\) surgery on a knot in \(S^3\) satisfy certain conditions; this gives an obstruction to unknotting number one which was used to great effect by \textit{P. S. Ozsváth} and \textit{Z. Szabó} in [Topology 44, No. 4, 705--745 (2005; Zbl 1083.57013)]. Greene combines these two obstructions, giving a beautiful refinement of the Donaldson obstruction above which applies in the case that \(X\) is \textit{sharp}, which means its intersection form may be used to compute the correction terms of its boundary. For an alternating knot \(K\), the double cover \(X_K\) of the 4-ball along the black surface of an alternating diagram is such a sharp manifold. The resulting theorem is that if the double branched cover of an alternating knot is half-integer Dehn surgery on a knot, then the intersection lattice of \(X_K\) is a \textit{changemaker lattice}. This means that it is isomorphic to the orthogonal complement in \(\mathbb{Z}^{m+2}\) of the span of vectors \(v=(1,-1,0,\dots,0)\) and \(w=(0,1,\sigma_1,\dots,\sigma_m)\); moreover the coefficients \(\sigma_1,\dots,\sigma_m\) of \(w\) satisfy the changemaker condition: every number between 1 and their sum is equal to the sum of some subset of \(\{\sigma_1,\dots,\sigma_m\}\). This turns out to be a very strong and useful condition. As an application, Greene classifies alternating 3-braid closures with unknotting number one. He asks whether his theorem may be used to show that every alternating knot with unknotting number one has an unknotting crossing in each alternating diagram. This has subsequently been answered in the affirmative by \textit{D. McCoy} [Adv. Math. 305, 757--802 (2017; Zbl 1353.57013)].