Singularities appearing on generic fibers of morphisms between smooth schemes (Q2519152)

The goal of the article is to explore the structure of singularities that occur in generic fibres in positive characteristic. The author determines which rational double points do and which do not occur on generic fibres. Let \(k\) be an algebraically closed ground field of characteristic \(p>0\), and suppose \(f:S \rightarrow B\) is a morphism between smooth integral schemes. Then the generic fiber \(S_\eta\) is a regular scheme of finite type over the function field \(E=\kappa(\eta).\) In characteristic \(0,\) this implies that \(S_\eta\) is smooth over \(E\). The absolute Galois group \(G=\operatorname{Gal}(\bar E/E)\) acts on the geometric generic fiber \(S_{\bar\eta}\) with quotient isomorphic to \(S_\eta\), so to understand the generic fiber it suffices to understand the geometric generic fiber which is again smooth over an algebraically closed field, together with its Galois action. The situation is more complicated in characteristic \(p>0\). The reason is that over nonperfect fields the notion of regularity is weaker than the notion of geometric regularity, which coincides with formal smoothness. Here it easily happens that the geometric generic fiber \(S_{\bar\eta}\) acquires singularities. As is proved by Bombieri and Mumford, there are quasi-elliptic fibrations of \(p=2\) and \(p=3\), which are analogous to elliptic fibrations but have a cusp on the geometric generic fiber. A proper morphism \(f:S\rightarrow B\) of smooth algebraic schemes is called a \textit{quasifibration} if \(\mathcal O_B=f_\ast(\mathcal O_S)\) and if the generic fiber \(S_\eta\) is not smooth. Quasi-fibrations involve some fascinating geometry and offer new freedom to achieve geometrical constructions that are impossible in characteristic \(0\). Nonsmoothness of the generic fiber \(S_\eta\) leads to unusual complications. However, singularities appearing on the geometric generic fiber \(S_{\bar\eta}\) are not arbitrary. First, they are locally of complete intersection; hence many powerful methods from commutative algebra apply. They also satisfy far more restrictive conditions, and the goal of the paper is to analyze these. Hirokado started an analysis, characterizing those rational double points in odd characteristic that appear on geometric fibres. His approach was to study the closed fibres \(S_b\) \((b\in B)\) and their deformation theory. In this article, the author looks at the generic fibre \(S_\eta\) and work over the function field \(\kappa(\eta).\) The author works in the following abstract setting: Given a field \(F\) in characteristic \(p>0\) and a subfield \(E\) such that the field extension \(E\subset F\) is purely inseparable. Then the author considers \(F\)-schemes \(X\) of finite type that descend to regular \(E\)-schemes \(Y\), that is, \(X\backsimeq Y\otimes_E F.\) The first results of such schemes are: In codimension 2, the local fundamental groups are trivial and the torsion of the local class groups are \(p\)-groups. Moreover, the Tjurina numbers are divisible by \(p\), the stalks of the Jacobian ideal have finite projective dimension, and the tangent sheaf \(\Theta_X\) is locally free in codimension \(2\). These conditions give strong conditions on the singularities. The first main result of the article is a surprising restriction on the cotangent sheaf: If an \(F\)-scheme \(X\) descends to a regular scheme then, for each point \(x\in X\) of codimension \(2\), the stalk \(\Omega^1_{X/F,x},\) contains an invertible direct summand. As an application of these results the author determines which rational double points appear on surfaces descending to regular schemes and which do not. It turns out that the situation is most challenging in characteristic 2: Besides the \(A_n\)-singularities, which behave as in characteristic \(0\), there are the following isomorphism classes: \(D_n^r\) with \(0\leq r\leq \lfloor n/2\rfloor-1\) and \(E_6^0, E_6^1,E_7^0,\dots,E_7^3,E_8^0,\dots,E_8^4.\) Notice that all the members of the list have a tangent sheaf that are locally free, although there are other rational double points with locally free tangent sheaf. The author sets up notation and gives some elementary examples and results. He analyzes \(F\)-schemes that \(X\) descend to regular \(E\)-schemes \(Y\), and treats the local fundamental groups. He proves that integer-valued invariants are multiples of \(p\), and he treats the finite projective dimension of sheaves obtained from the cotangent sheaf \(\Omega^1_{X/F}.\) This article is written in an easy to understand language, it is more or less self contained with good references when needed. Finally, it treats a computable field of singularity theory which is of importance to the study in positive characteristic.