Toward resolution of singularities over a field of positive characteristic (The idealistic filtration program). II: Basic invariants associated to the idealistic filtration and their properties (Q983318)

A few years ago H. Kawanoue announced a program (the \textit{Idealistic Filtration Program}) to resolve singularities of algebraic varieties which might work even if the base field is of positive characteristic. It is expressed in terms of \textit{idealistic filtrations}. If \(U\) is an open set of an affine variety \(W\), with coordinate ring \(R\), an idealistic filtration on \(W\) (or on \(R\)) is an indexed family of ideals \(I_a\) of \(R\), where \(a\) is a non-negative real number, such that \(I_a \subset I_b\) if \(a \geq b\), \(I_a I_b \subset I_{a+b}\) and \(I_0=R\). One defines the support, or singular locus, of an idealistic filtration. Then the goal is to give an algorithm allowing us to resolve the filtration, i.e., to find blowing-ups with smooth centres such that, taking suitable transforms, eventually the support becomes empty. The centres will be given as the set of points where certain \textit{algorithmic resolution functions}, with values in a suitable totally ordered set, reach the maximum value. From this, a similar algorithm to resolve singularities of varieties will follow. To construct such resolutions functions, given an idealistic filtration \(\mathcal I\) on the variety \(W\) a \textit{strand of invariants} is attached to each point \(P\) of \(W\). Each step of the strand involves a triplet \(\sigma\), \(\tilde {\mu}\) and \(s\) of invariants. In the definition of \(\sigma\) and \(\tilde \mu\) a fundamental role is played by \(L(\mathcal I)\), a certain graded sub-algebra of \(Gr_{M}(A)\), the graded ring of \((A,M)\) (the local ring of \(W\) at a closed point \(P\)). Of particular importance is the notion of \textit{leading generator system}. These are elements in suitable ideals \(I_j\) of \(\mathcal I\) whose initial forms in the graded ring \(Gr_{M}(A)\) are generators of \(L(\mathcal I)\), satisfying certain ``minimality'' conditions. This notion seems essential to avoid the explicit use of hypersurfaces of maximal contact, a usual tool in desingularization techniques which is not always available in positive characteristic. Leading generator systems exist when the filtration \(\mathcal I\) is \(D\)-saturated, that is closed under the application of differential operators. The resolution functions thus obtained have the fundamental property necessary in problems of this type. Namely, their values strictly decrease after performing a blowing-up with the appropriate algorithmic center. But still, when the characteristic is positive, there is no warranty that the process terminates. Kawanoue, working in cooperation with K. Matsuki, plans to present the details of the program in a series of four papers. In the first one, \textit{H. Kawanoue} [Publ. Res. Inst. Math. Sci. 43, 819--909 (2007; Zbl 1170.14012)], discussed the main lines of the program, in particular what was mentioned so far. In the second part of the series (the article under review) the authors carefully study properties of the invariants \(\sigma\) and \(\tilde \mu\) (at ``time zero'', i.e., at the beginning of the process). For instance, they check the upper-semicontinuity of the invariants \(\sigma\) and \(\tilde \mu\). They also discuss certain power series expansions, but nor necessarily in terms of a regular system of parameters, but rather relative to a leading generating system; something that will be important later. Many of these results had been announced (without proof) in Part I. In an Appendix, they include an important improvement of a result in Part I. Over there, they prove what they call \textit{the new non-singularity principle}. It says (among other things) that given an idealistic filtration \(\mathcal I\) which is both \(D\)-saturated and \(R\)-saturated (i.e., closed under the operations of differentiation and extraction of roots) and satisfies, for some closed point, \(\tilde \mu (z) = \infty\), then on a neighbourhood of \(z\) the support of \(\mathcal I\) is non-singular. From this result it follows that the centres involved in the application of the algorithm are indeed regular, which is a non-trivial result. In this Appendix a similar result is proved, but avoiding the hypothesis of \(R\)-saturation. This improvement could be an important step toward proving that their algorithm (or a slight variation thereof) will terminate after a finite number of steps, thus concluding a proof of resolution in positive characteristic. Indeed, the assumption on \(R\)-saturation seems to be an obstacle to prove the termination of the algorithm. A subsection (numbered 0.3) resumes the current status of the Idealistic Filtration Program.