Looijenga's conjecture via integral-affine geometry (Q1657718)

The main result of this article is a new proof of Looijenga's conjecture on cusp singularities. A cusp singularity \((\overline{V},p)\) is a germ of surface singularity such that the exceptional divisor \(D\) of the minimal resolution \(\pi: V\to \overline{V}\) is a reduced cycle \(\pi^{-1}(p)= D=D_1+\cdots+D_n\in |-K_V|\) of smooth rational curves \(D_i\) meeting transversely. It is known that each cusp surface singularity \((\overline{V},p)\) has its \textit{dual} cusp singularity \((\overline{V}',p')\). One of the constructions relating \((\overline{V},p)\) and \((\overline{V}',p')\) is via \textit{hyperbolic Inoue surface}, and it is recalled in the paper. \textit{E. Looijenga} [Ann. Math. (2) 114, 267--322 (1981; Zbl 0509.14035)] proved that whenever a cusp singularity \((\overline{V},p)\) is smoothable, the exceptional divisor \(D\) of the minimal resolution of its dual cusp singularity \((\overline{V}',p')\) (\(D'\) is then called dual to \(D\)) is an anticanonical divisor of some smooth rational surface \(Y'\). Looijenga also conjectured the converse, i.e., that existence of such an \textit{anticanonical pair} \((Y',D')\) implies the smoothability of the cusp singularity with cycle \(D\). The first proof of Looijenga's conjecture was obtained by \textit{M. Gross} et al. [Publ. Math., Inst. Hautes Étud. Sci. 122, 65--168 (2015; Zbl 1351.14024)], as an application of mirror symmetry. In the present paper, P. Engel gives an alternative proof of Looijenga's conjecture. The new proof does not make any explicit use of mirror symmetry. Instead, it relies mainly on Friedman-Miranda's criterion of smoothability of cusps with cycle \(D\), see [\textit{R. Friedman} and \textit{R. Miranda} [Math. Ann. 263, 185--212 (1983; Zbl 0488.14006)], and on the theory of integral-affine structures on real surfaces. The work of Friedman and Miranda is quoted in Gross, Hacking, and Keel's paper, but seems not to be used. The scheme of the proof is, roughly, as follows. First, an anticanonical pair \((Y',D')\) is represented as a sequence of blow-ups and smoothings of a toric surface \((\overline{Y}',\overline{D}')\). The moment polygon \(\overline{S}'\) of \((\overline{Y}',\overline{D}')\) is endowed with an appropriate integral-affine structure and then it admits a sequence of surgeries parallel to that of \((\overline{Y}',\overline{D}')\). In this way, a new integral-affine real surface \(S\) is produced. After completing \(S\) to a topological triangulated sphere \(\hat{S}\) and passing to a certain refinement \(\hat{S}[k]\), it is shown that to each vertex \(v_i\), \(i>0\), of the triangulation one can associate an anticanonical pair \((V_i,D_i)\), and there is also a special vertex \(v_0\) to which one associates the hyperbolic Inoue pair \((V_0,D)\), where \(D\) is the cycle dual to \(D'\). All the surfaces \(V_i\), \(i\geq 0\), glue well and form a reducible surface \(\mathcal{X}_0\) whose dual complex is \(\hat{S}[k]\) and which is smoothable by Friedman-Miranda's criterion. The proofs of Looijenga's conjecture by Gross, Hacking, and Keel and by Engel make an impression of being of the same spirit. At the same time, both are highly non-trivial and a direct connection between them is hard to specify. Some comments on this are given by P. Engel in the introduction to his paper.