Cyclic and finite surgeries on Montesinos knots (Q1017848)

It is well known that there are two pretzel knots which admit cyclic surgery or finite surgery. One is the \((-2,3,7)\)-pretzel knot and the other one the \((-2,3,9)\)-pretzel knot. For the former, the 18 and 19-surgeries yield lens spaces, and the 17-surgery yields a Seifert fibered manifold with finite non-cyclic fundamental group. For the latter, the 22 and 23-surgeries yield such Seifert fibered manifolds. The main result of the paper under review claims that these are the only non-trivial cyclic or finite non-cyclic Dehn surgeries on hyperbolic Montesinos knots. Combined with a result of \textit{Y.-Q. Wu} [J. Differ. Geom. 43, No.~1, 171--197 (1996; Zbl 0851.57018)], the same conclusion is true among hyperbolic arborescent knots. The argument is summarized as follows. Assume a hyperbolic Montesinos knot \(K\) admits a non-trivial cyclic or finite non-cyclic Dehn surgery. \textit{Charles Delman} [Constructing essential laminations which survive all Dehn surgeries, unpublished paper] showed that the knot \(K\) is either a \((-2\ell,p,q)\)-pretzel knot, or a \((-1,2n,p,q)\)-pretzel knot, or a \((-1,-1,2m,p,q)\)-pretzel knot with an integer \(n\), integers \(\ell\), \(m>1\), and odd integers \(p\), \(q\), \(3\leq p\leq q\). The first class was excluded by \textit{T. W. Mattman} [J. Knot Theory Ramifications 11, No.~6, 891--902 (2002; Zbl 1023.57016)], and the third is excluded by showing that the knots are not fibered by using the algorithm of \textit{D. Gabai} [Comment. Math. Helv. 61, 519--555 (1986; Zbl 0621.57003)]. For, \textit{Y. Ni} [Invent. Math. 170, No. 3, 577--608 (2007; Zbl 1138.57031)] showed that only fibered knots admit \(L\)-space surgery, including cyclic surgery and finite surgery. Thus the second class remains. By direct calculations, the Alexander polynomial of \(K\) has a coefficient not equal to \(\pm 1\), unless \(n=1\) and \(p=3\). On the other hand, \textit{P. Ozsváth} and \textit{Z. Szabó} [Topology 44, No. 6, 1281--1300 (2005; Zbl 1077.57012)] showed that if a knot admits an integral Dehn surgery yielding an \(L\)-space then its Alexander polynomial has only \(\pm 1\) as its coefficient. By the famous cyclic surgery theorem by \textit{M. Culler, C. McA. Gordon, J. Luecke} and \textit{P. B. Shalen} [Ann. Math. (2) 125, 237--300 (1987; Zbl 0633.57006)], any cyclic surgery on a hyperbolic knot must be integral. However, a finite surgery is not shown to be integral yet. Hence, the authors need to show that the existence of a non-integral \(L\)-space surgery implies that of an integral \(L\)-space surgery by calculating some Heegaard Floer homology. Thus a \((-1,2,3,q)\)-pretzel knot with odd \(q\geq 3\) is the only candidate. This is equivalent to a \((-2,3,q)\)-pretzel knot. For this type of knot, \textit{T. W. Mattman} [J. Knot Theory Ramifications 11, No.~6, 891--902 (2002; Zbl 1023.57016)] has already obtained the desired conclusion. Independently, \textit{D. Futer, M. Ishikawa, Y. Kabaya, T. W. Mattman} and \textit{K. Shimokawa} [Algebr. Geom. Topol. 9, No. 2, 743--771 (2009; Zbl 1166.57003)] proved that a \((-2,p,q)\)-pretzel knot for odd \(p,q\geq 5\) has neither non-trivial cyclic nor non-cyclic finite Dehn surgery by using Culler-Shalen theory.