Wild singularities of flat surfaces (Q375758)

A flat surface is a pair \((X, \omega)\) of a Riemann surface \(X\) and a holomorphic \(1\)-form \(\omega\not=0\) on \(X\). Let \(Z(\omega)\) be the set of zeros of \(\omega\) and \(X'=X\setminus Z(\omega)\). Then \(\omega\) defines a flat Riemannian metric on \(X'\). Let \(\hat{X}\) be the metric completion of \(X\). Throughout the paper it is assumed that \(\text{Sing}(X)=\hat{X}\setminus X'\) is a discrete subset of \(\hat{X}\). The points of \(\text{Sing}(X)\) are classified into flat points, finite angle singularities, infinite angle singularities and \textit{wild} singularities. The paper under review emphasizes the analysis of a topological space \({\mathcal L}(x)\) which the authors call the set of linear approaches to the point \(x\). This set \({\mathcal L}(x)\) consists of equivalence classes of unit speed geodesic trajectories \(\gamma : (0,\epsilon_{\gamma})\to X'\) such that \(\lim_{t\to +0}\gamma(t)=x\). Such geodesic trajectories \(\gamma_1\) and \(\gamma_2\) are equivalent if \(\gamma_1(t)=\gamma_2(t)\) for all \(t\in(0, \min\{\epsilon_{\gamma_1},\epsilon_{\gamma_2}\})\). The topological structure of \({\mathcal L}(x)\) is studied in detail in Section 2 and proved to be an extension of the unit tangent bundle of \(X\) to \(\hat{X}\) in Proposition 2.12. It is also shown in Corollary 3.4 that the homeomorphism type of \({\mathcal L}(x)\) is an affine invariant. The set \({\mathcal L}(x)\) is decomposed into \textit{rotational components}. An \textit{angular sector} is a triple \((I,c,i_c)\) of a non-empty interval (perhaps a one-point set) \(I\subseteq \mathbb{R}\), a real number \(c\) and an isometry \(i_c: U(I,c)\to X'\), where \(U(I,c)=\{z=x+iy :x<c, y\in I\}\) is equipped with the \(1\)-form \(e^zdz\). Then two points \([\gamma_1]\) and \([\gamma_2]\) belong to the same rotational component if there exists an angular segment \((I,c,i_c)\) such that some representatives \(\gamma_1\) and \(\gamma_2\) are chosen so that \((i_c^{-1}\circ \gamma_i)(0,\epsilon_{\gamma_i})\) are full horizontal rays in \(U(I,c)\). If \(x\) is not wild, then \({\mathcal L}(x)\) has only one rotational component and it is isometric to \(\mathbb{R}/G\) with \(G\) a subgroup of \(\mathbb{Z}\). For a wild singularity, there are many possible cases for the numbers and topological types of rotation components, see the examples in Section 1.3. The authors define functions \(\text{dir}\), \(\ell_{\epsilon}\) and \(\sigma_{\epsilon}\) for \(\epsilon>0\) on \({\mathcal L}(x)\) and transverse measures \(\mu_{\theta}\) on Borel subsets of \({\mathcal L}_{\theta}(x)\) for \(\theta\in S^1\). Here \({\mathcal L}_{\theta}(x)\) is the subset of linear approaches in \({\mathcal L}(x)\) having the direction \(\theta\). Let \(X\) and \(Y\) be translation surfaces. Let \(x\) and \(y\) be points in \(\hat{X}\) and \(\hat{Y}\), respectively. Theorem 4.1 shows that the existence of a homeomorphism \(F :{\mathcal L}(x)\to {\mathcal L}(y)\) which preserves \(\ell_{\epsilon}\), \(\sigma_{\epsilon}\) for some \(\epsilon>0\), \(\text{dir}\) and \(\mu_{\theta}\) for all \(\theta\in S^1\) guarantees that \(x\) and \(y\) have isometric neighborhoods. In Section 5, the authors compare \({\mathcal L}(x)\) at a wild singularity \(x\) with the Alexandrov cone and the Gromov-Hausdorff tangent cone constructed by other metric completions and demonstrate an advantage of \({\mathcal L}(x)\) over the others. It seems that the researches by the authors are still ongoing. It is desirable that the results in this paper (in particular those in Section 3) will be reinforced by many examples.