L-space surgeries, genus bounds, and the cabling conjecture (Q2354902)

Let \(K\) be a knot with genus \(g(K)\). Suppose that a positive integral surgery \(p\) on \(K\) yields a lens space. P. Kronheimer, T. Mrowka, P. Ozsváth and Z. Szabó [\textit{P. B. Kronheimer} et al., Ann. Math. (2) 165, No. 2, 457--546 (2007; Zbl 1204.57038)] showed that \(2g(K)-1\leq p\) by using monopole Floer homology. In fact, this holds for \(L\)-space surgery. The main result gives a sharper inequality \(2g(K)-1\leq p-\sqrt{p}-1\) under the assumption that the resulting \(L\)-space bounds a smooth negative definite \(4\)-manifold \(X\) with torsion-free first homology. Moreover, if \(X\) can be chosen to satisfy another condition involving correction terms, then the improved bound \(2g(K)-1\leq p-\sqrt{3p+1}\) holds. The author gives an infinite family of knots, each of which attains the equality in the latter. As an application of the main result, a partial answer to the cabling conjecture is proved. If a surgery on a knot yields a connected sum of lens spaces, then the knot is either a torus knot or a cable of a torus knot. The surgery slope is also identified.