The number of equisingular moduli of a rational surface singularity (Q1692609)

This paper proves a conjectured topological inequality for the number of equisingular moduli of a rational surface singularity in the case that the self-intersections of the components of the exceptional divisor on the minimal good resolution are sufficiently negative. More precisely, let \((X,E)\longrightarrow(V,0)\) be the minimal good resolution of a complex normal surface singularity and let \(S_X=(\Omega^1_X(\log E))^*\) be the bundle of logarithmic derivations. The author conjectured [Algebr. Geom. 2, No. 3, 315--331 (2015; Zbl 1322.14056)] that for general surface singularities \[ h^1(\mathcal{O}_X) - h^1(S_X) + h^1(\bigwedge^2S_X) \geq 0, \] with equality if and only if \((V,0)\) is quasi-homogeneous (if \((V,0)\) is not Gorenstein, otherwise the inequality is true with 1 on the right hand side). For a rational singularity the term \( h^1(S_X)\) measures the number of equisingular moduli and is difficult to compute. The next term equals \(h^1(-(K_X+E))\), the second plurigenus, and can be computed from the resolution graph. In the same paper the Rational Conjecture (i.e., the rational case of the general conjecture) was shown for star-shaped graphs. The \(h^1\) of the restrictions of the bundles involved to the exceptional set can be computed solely from the graph. If \((V,0)\) is not simple elliptic or a cusp, then \(h^1(\mathcal{O}_E) - h^1(S_X\otimes\mathcal{O}_E) + h^1(-(K_X+E)\otimes\mathcal{O}_E) =1-\delta\) with \(\delta=1\) for a star-shaped graph and \(0\) otherwise. So the inequality in the Main Conjecture holds when \(h^1(S_X)=h^1(S_X\otimes\mathcal{O}_E)\), e.g. if \(h^1(S_X(-E)) = 0\). If the \(-E_i\cdot E_i\) are big enough relative to the genus and valency of \(E_i\) for all \(i\), then such a cohomological vanishing result can be proved via so-called ``easy vanishing theorems'' of the author [Invent. Math. 31, 17--41 (1975; Zbl 0314.14010)]. To prove the Rational Conjecture in more cases, with weaker assumptions on \(-E_i\cdot E_i\) a ``hard vanishing theorem'' is needed, which is false in prime characteristic. The long argument involves a detailed study of maps between cohomology groups. To formulate one result, let \(t_i'\) denote the number of intersections with curves which are not end-curves. If for a rational singularity with reduced fundamental cycle \(-E_i\cdot E_i\geq t_i+t_i'-2\) for all \(i\), except for exactly one or exactly two separated by a string of rational curves \(-E_i\cdot E_i= t_i+t_i'-3\), then the Rational Conjecture holds. The computations allow to recover easily a part of Laufer's classification of taut singularities, namely those with reduced fundamental cycle. The methods of this paper enable one to extend this partial classification to characteristic \(p\).