Resolution of singularities of the cotangent sheaf of a singular variety (Q502660)

This paper is devoted to the problem of desingularization of the cotangent sheaf of algebraic varieties and analytic spaces. To be more precise, let \(X_0\) be a reduced algebraic variety over a field of characteristic zero or a complex or real analytic space, and denote by \(\mathrm{Sing } X_0\) the singular locus of \(X_0\). A resolution of singularities of \(X_0\) is, as usual, a proper birational or bimeromorphic morphism \(\sigma: X\to X_0\) from a smooth \(X\) such that \(\sigma\) is an isomorphism over \(X_0\setminus\mathrm{Sing } X_0\) and \(\sigma^{-1}(\mathrm{Sing } X_0)\) is a divisor with simple normal crossings on \(X\). Often one requires also that \(\sigma\) is a composite of blowings-up with smooth admissible (i.e., possessing only normal crossings with exceptional divisor) centers. The main result of the paper states that, for \(X_0\) as above and of dimension \(\dim X_0\leq 3\), there exists a resolution of singularities which is a composite of blowings-up with smooth admissible centers such that the pull-back of the cotangent sheaf of \(X_0\) is locally generated by differential monomials \(d\,(\mathbf{u}^{\mathbf{\alpha}_i})\), \(i=1,\ldots,s\), and \(d\,(\mathbf{u}^{\mathbf{\beta}_j}v_j)\), \(j=1,\ldots,n-s\), where \(n=\dim X_0\), \((u_1,\ldots,u_s,v_1,\ldots,v_{n-s})\) is a local coordinate system on \(X\) such that (1) the support of the exceptional divisor is locally defined by the equation \(u_1\cdots u_s=0\), (2) the multiindices \(\mathbf{\alpha}_1,\ldots,\mathbf{\alpha}_s\in\mathbb{N}^s\) are linearly independent over \(\mathbb{Q}\), and (3) the set \(\{\mathbf{\alpha}_i,\mathbf{\beta}_j\}\) is totally ordered with respect to the componentwise partial ordering of \(\mathbb{N}^s\). The approach undertaken by the authors of this paper consists in showing that resolution of the cotangent sheaf of a variety \(X_0\) is equivalent to principalization of the logarithmic Fitting ideals \(\mathcal{F}_k(\sigma)\) of the morphism \(\sigma\). These ideals are generated locally by the minors of order \(n-k\) of the logarithmic Jacobian matrix of \(\sigma\) (see Subsection~2.2 of the paper for a precise definition). A resolution \(\sigma: X\to X_0\) is a principalization of the logarithmic Fitting ideals if all \(\mathcal{F}_k(\sigma)\), \(k=0,\ldots,n-1\), are principal and generated locally by monomials in components of the exceptional divisor. The authors prove a general theorem that a principalization of \(\mathcal{F}_k(\sigma)\) at a point \(a\in X\) where the logarithmic Jacobian matrix has rank \(r\) can be achieved by standard resolution technique for \(k=n-1,\ldots,n-r-1\). In dimension \(3\), the only case that remains is to principalize \(\mathcal{F}_1(\sigma)\) at points of logarithmic rank \(0\). The proof in this case is the most difficult part of the paper and perhaps this is the point that cannot be easily generalized to higher dimensions. As in other variants of resolution theorems, the authors construct a technical nonnegative integral invariant that drops after a suitable admissible blowing-up and this allows to proceed by induction. In dimensions \(\dim X_0\geq 4\), the assertion of the main theorem remains a conjecture.